a b s t r a c t

Objective: To compare the association of 3-h sepsis bundle compliance with hospital mortality in non-hypoten- sive sepsis patients with intermediate versus severe hyperlactemia.

Methods: This was a cohort study of all non-hypotensive, hyperlactemic sepsis patients captured in a prospective quality-improvement database, treated October 2014 to September 2015 at five tertiary-care centers. We defined sepsis as 1) infection, 2) >=2 SIRS criteria, and 3) >=1 organ dysfunction criterion. “Time-zero” was the first time a

patient met all Sepsis criteria. Inclusion criteria: systolic blood pressure N 90 mmHg, mean arterial pressure N 65

mmHg, and serum lactate >=2.2 mmol/L. Primary exposures: 1) intermediate(2.2-3.9 mmol/L) versus severe(>=4.0 mmol/L) hyperlactemia and 2) full 3-h bundle compliance. Bundle elements:

Blood cultures before antibiotics.

Parenteral antibiotics administered <= 180 min from >=2 SIRS and lactate ordered, or <= 60 min from "time-zero", whichever occurred first.

Lactate result available <=90 min post-order.

30 mL/kg crystalloid bolus initiated <=30 min from "time-zero".

The primary outcome was 60-day in-hospital mortality.

Results: 2417 patients met inclusion criteria. 704(29%) had lactate >= 4.0 mmol/L versus 1775 patients with lactate

2.2-3.9 mmol/L. Compliance was 75% for antibiotics and 53% for fluids. Full-compliance was comparable between lactate groups (n = 200(29%) and 488(28%), respectively). We observed 424(17.5%) mortalities: intermediate/ non-compliant - 182(14.9%), intermediate/compliant - 41(8.4%), severe/non-compliant - 147(29.2%), severe/ compliant - 54(27.0%) [difference-of-differences = 4.3%, CI = 2.6-5.9%]. In multivariable regression, Mortality predictors included severe hyperlactemia (OR = 1.99, CI = 1.51-2.63) and bundle compliance (OR = 0.62, CI = 0.42-0.90), and their interaction was significant: p_{(interaction)} = 0.022.

Conclusion: We observed a significant interaction between 3-h bundle compliance and initial hyperlactemia. Bundle compliance may be associated with greater Mortality benefit for non-hypotensive sepsis patients with less severe hyperlactemia.

(C) 2017

Introduction

* Corresponding author at: Department of Emergency Medicine, Northwell Health System, 300 Community Drive, Manhasset, NY 11030, United States.

E-mail address: [email protected] (D.E. Leisman).

Background

Sepsis and septic shock are global drivers of mortality, estimated to account for 5 million deaths annually [1]. An extensive body of literature

http://dx.doi.org/10.1016/j.ajem.2017.01.029

0735-6757/(C) 2017

has demonstrated compliance with early-action Sepsis bundles, man- dating immediate intravenous fluid resuscitation and empiric broad- spectrum antimicrobial therapy, to be associated with reduced hospital mortality for these patients [2-8]. Hyperlactemia is considered by many to be an important indicator of sepsis-induced tissue hypoperfusion and has also been consistently shown to predict sepsis mortality [9-15].

The quantitative threshold for lactate elevation as a bundle trigger, or “time-zero” entry point, has been the subject of debate for 25 years. The first sepsis consensus definitions (Sepsis-1) in 1991 recommended lactate >= 4.0 mmol/L as an indicator of septic shock and a time-zero entry point [16]. While the 2001 Sepsis-2 definitions lowered the lactate

threshold for sepsis-induced tissue hypoperfusion to 3.0 mmol/L [17], subsequent Society for Critical Care Medicine (SCCM) and National Quality Forum (NQF) recommendations continued to support 4.0 mmol/L as cut-off for bundle triggers [18,19]. During that time, a number of studies suggested patients with more intermediate hyperlactemia, with serum levels 2.1-4.0 mmol/L, are also at increased risk for decompensation and mortality, and may benefit from early sep- sis bundle application [11-13,20]. The recent SCCM and European Soci- ety for Intensive Care Medicine (ESICM) Sepsis-3 definitions include

lactate level >= 2.1 mmol/L to clinically identify septic shock [21].

Importance

These new definitions acknowledge intermediate hyperlactemia as a risk-factor for sepsis mortality [21], and analyses that accompany the definitions compare severe (>= 4.0 mmol/L) vs. intermediate (N 2.0 mmol/L) hyperlactemia as predictors of hospital death [13]. However, the differential effect of initial sepsis bundle compliance (i.e., initial

treatment) in these groups has not been well investigated. While sepsis may be viewed as a question of whether a patient is or is not septic, the dysregulated host response of sepsis may be seen as a continuum where patients experience progressively worsening organ dysfunction [21]. As Inclusion/exclusion criteria“>a result, it is conceivable that patients with earlier or less severe hypo- perfusion, (and specifically with intermediate versus severe hyperlactemia), might be at an earlier point along such a continuum, and could therefore demonstrate a comparatively greater mortality benefit from bundle compliant initial care. If true, these patients would represent a high-impact target population. Such knowledge would further our understanding of the best practice in managing sepsis patients.

Study goal

We conducted an analysis of a hemodynamically stable sepsis co- hort, identified from a prospective, multisite quality improvement (QI) sepsis database. We aimed to determine the differential 60-day in-hospital mortality benefit associated with bundle compliance for non-hypotensive sepsis patients presenting with intermediate hyperlactemia compared to those with severe hyperlactemia. As a sec- ondary analysis, we also assessed 28-day in-hospital mortality and need for mechanical ventilation.

Methods

This was a multisite, observational cohort study, examining data from a sepsis quality improvement (QI) database operating across 11 hospitals in a single U.S. health system. The health system began a sepsis QI program in 2009 and adopted a defined algorithm and three-hour bundle in 2010 to screen and treat sepsis patients [22]. To measure bun- dle compliance and outcomes for quality and research purposes, data for all consecutive sepsis and septic shock patients were prospectively captured in an internally managed QI database. We abstracted records for hyperlactemic, non-hypotensive sepsis patients treated in 2015 at any of the health system’s 5 tertiary hospitals into a separate research

registry to compare the association of bundle compliance with in-hospi- tal mortality in patients with intermediate versus severe hyperlactemia.

Database methods

A dedicated team of data abstractors at each site screened all pa- tients with known or suspected infection and >= 2 SIRS criteria for data- base capture. Abstractors only recorded information for patients meeting database inclusion criteria: 1) a suspected or confirmed infec- tion, 2) >= 2 SIRS criteria [17], and 3) lactate >= 2.2 mmol/L or hypotension

(systolic blood pressure b 90 mmHg or mean arterial pressure b 65

mmHg) or >= 1 sepsis organ-dysfunction criteria (outlined below). Rele- vant data were abstracted into the QI database using a standardized data collection form, which abstractors submitted to a centralized data

collection unit. Abstractors excluded patients: b 18 years; with Advance directives precluding bundle interventions; who declined interven- tions; who were admitted from the ED directly to palliative care or hos- pice; or who were enrolled in an IRB-approved clinical trial that precluded standard application of the bundle. Demographic and clinical data obtained included patient age, sex, primary payer, initial lactate level, signs of hypoperfusion or organ dysfunction, and comorbidities at baseline, as well as treatment and laboratory data-points. Database managers monitored database quality monthly. Quality control was ap- plied by monthly gap-analyses between QI database records and a data- base of all hospital discharges administered by New York State [23]. Records for patients with a discharge diagnosis of sepsis who were missing from the database were reviewed to determine if database in- clusion criteria were met, and entered if appropriate. Patients who were “missed” by the treating clinician would have had their initial sep- sis episode entered as the first time all objective inclusion criteria were met based on manual review of the medical record. All abstractors re- ceived standardized training at the beginning of data-collection involvement.

Study inclusion/exclusion criteria

The inclusion criterion for this study was treatment in a tertiary care facility and capture in the QI database (i.e., confirmed or suspected in- fection, >= 2 SIRS, and organ dysfunction or lactate >= 2.2 mmol/L). We ex- cluded all database patients who initially presented with hypotension (systolic blood pressure b 90 mmHg or mean arterial pressure b 65 mmHg) and who did not have a lactate level >= 2.2 mmol/L. We excluded patients with lactate b 2.2 rather than b 2.1 because at the time of algo- rithm development, lactate <= 2.1 mmol/L was within the error-margin

of normal for the sites’ laboratories and not collected: the QI database

inclusion criteria require >=1 hypoperfusion or organ-dysfunction crite- rion be met and use >=2.2 as the inclusion threshold for lactate.

Study definitions

Sepsis was defined as 1) infection, 2) >= 2 SIRS criteria, and 3) lac- tate >= 2.2 mmol/L or acute organ-dysfunction (not otherwise explained by the patient’s past medical history). Organ-dysfunction criteria were:

Acute kidney injury , coagulopathy, hypoxia, elevated bilirubin (>= 2.0 mg/dL), and altered mentation. AKI was serum creatinine N 2.0 mg/dL in the absence of chronic kidney disease or 50% increase from known baseline. Coagulopathy was platelet count b 150 000 cells/um³, International normalized ratio N 1.5,

activated partial thromboplastin time N 30 s, or partial thromboplastin time N 60 s. Hypoxia was a new, increased oxygen requirement to main- tain SaO₂ N 90% or a PaO₂/FiO₂ ratio b 300. Altered mentation was an acute change in mental status determined by clinical judgment of the treating physician. To expedite algorithm inclusion, locally developed consensus-criteria, nicknamed ‘Super-SIRS’ (included in Table 1), were

employed as an additional time-zero entry-point if >= 2 criteria were met at ED triage. In this investigation, all patients had lactate >= 2.2

Table 1

Organ dysfunction “Time-Zero” criteria definitions.

mmol/L and b 4.0 mmol/L, and severe hyperlactemia as a lactate >= 4.0 mmol/L.

Organ dysfunction criteria^a

1. Acute Kidney Injury ^b

Definition

Serum creatinine N 2.0 mg/dL or 50% increase from known baseline in the absence of chronic kidney disease

Study data

For this study, we extracted data for all eligible encounters in the QI

Thrombocytopenia Platelet count b 150 000 cells/um3

Coagulopathy^c International normalized ratio N 1.5, activated

partial thromboplastin time N 30 s, or partial thromboplastin time N 60 s, not otherwise explained by medical history

Elevated bilirubin Serum bilirubin N 2.0 mg/dL in the absence of pre-

existing Liver failure

Acute altered mental New altered mentation unrelated to the patient’s prior status (AMS) medical history

Hypoxia^c New increased O₂ requirement to maintain SaO₂ N 90% or a PaO₂/FiO₂ ratio b 300

database into a separate, IRB-approved research registry. Demographic variables extracted included age, sex, and body mass index (BMI). Clin- ical data included comorbidity status at presentation (congestive heart failure (CHF), chronic renal failure (CRF), chronic obstructive pulmo- nary disease (COPD), liver failure, immune modifying medications (IMMs), metastatic disease, and leukemia, lymphoma, or multiple mye- loma), site of infection (e.g., respiratory, urinary, etc.), suspected noso- comial etiology, whether the patient met ‘Super-SIRS’ criteria at triage, whether chest radiography indicated a lower respiratory tract infection

>= 2 “Super-SIRS”

criteria at triage

Locally developed consensus-criteria, where

meeting >= 2 criteria at triage was a “time-zero” entry point for 3-h bundle care. “Super-SIRS” criteria were:

Heart rate greater >= 120

Respiratory rate >= 24

Systolic blood pressure b 90 mmHg or 40% decrease from known baseline or mean arterial

pressure b 65 mmHg^d

Temperature >= 38.0 ?C (101.0? F) or <= 36.0 ?C (96.8? F)

Acutely altered mental status

at the time of the initial sepsis episode, initial lactate, and hypoperfusion or organ-dysfunction criteria at presentation.

The exposures of interest were intermediate vs. severe hyperlactemia at presentation and 3-h bundle compliance. Treatment data included times of intravenous fluid resuscitation initiation, lactate order and result, blood culture collection, and parenteral broad-spec- trum antibiotic administration, as well as ordered fluid volumes. De- tailed methods of capture and calculation for these measures have been described [24].

Outcomes

^a These 7 “time-zero” triggers were used as 3-h bundle entry points as well as study inclusion criteria. The intention was for all patients with a suspected infection who met >=2 SIRS criteria and any one of these criteria to receive carefully adherent to all 3-h bundle el- ements. All patients included in this study had a source of infection, met at least 2 SIRS criteria, and met at least one of the above organ dysfunction criteria. Time-zero was the first laboratory result or vital sign measurement time that caused the patient to meet

any of the above criteria.

^b Adapted from Kidney Disease|Improving Global Outcomes (KDIGO) criteria for de-

fining acute kidney injury.

^c Adapted from the 2001 International Sepsis Definitions Conference (Sepsis-2) report.

^d Any patient having met Super-SIRS criteria because of hypotension would have been excluded.

mmol/L, and we report the distribution of these additional organ dys- function measures within the study population.

We defined time-zero as the first laboratory result time or vital sign measurement time where a patient met all inclusion criteria: i.e., the first time a patient had 1) confirmed or suspected infection, and 2) >= 2

SIRS, and 3) >= 1 organ dysfunction criterion or lactate >= 2.2 mmol/L or

>= 2 “Super-SIRS” criteria (see Table 1). We select this point as “time- zero” because it is the first time that the treating physician had enough

information available to confirm that the patient was eligible for and should have received care adherent to the study sites’ algorithm and 3-h bundle. All sepsis patients, as defined above, met eligibility for the algorithm and 3-h bundle (detailed algorithm description provided in Fig. 1S). 3-h bundle elements were:

blood cultures drawn prior to antibiotic administration

Source-directed, broad-spectrum, parenteral antibiotics adminis- tered within 180 min of sepsis identification (i.e., >= 2 SIRS and a lac- tate ordered) or 60 min of time-zero (i.e., >= 2 SIRS and available laboratory results or vital signs indicating hypoperfusion or organ-

dysfunction), whichever occurs earlier.

Lactate result available within 90 min of order (ordered upon recog- nition of infection with SIRS)

30 mL/kg intravenous crystalloid bolus initiated within 30 min of time-zero We define bundle compliance as accomplishment of all 4 bundle el-

ements, and non-compliance as failure to achieve >= 1 bundle element.

The intention was to provide bundle compliant care to all eligible pa- tients. Care beyond 3-h was not protocolized and at the physician’s dis- cretion. We define intermediate hyperlactemia as an initial lactate >= 2.2

The primary outcome was 60-day in-hospital mortality. We also re- port 28-day in-hospital mortality. We assessed need for mechanical ventilation (defined as either: via endotracheal intubation or bi-level positive airway pressure) as a secondary outcome. In the context of the multifactorial nature of septic disease and the ‘composite’ nature of mortality as a sepsis outcome [25], we reasoned mechanical ventila- tion to be an appropriate secondary outcome that was captured in the database and more temporally proximal to the exposures assessed than mortality. We also considered Hemodynamic collapse requiring vasopressor administration as a secondary outcome, but low event rates, particularly in the compliant groups, precluded this analysis. Other measures of organ support that would have been candidates for secondary outcomes could not be obtained from the database (e.g., renal replacement therapy utilization).

Statistical analysis

We report continuous variables as means (standard deviations) or medians (interquartile ranges), and categorical variables as proportions. We constructed 95% confidence intervals and performed analyses with SPSS version 24.0 (IBM, Armonk, NY).

To assess differential mortality, we constructed Multivariable logistic regression models that adjusted for variables we anticipated a priori to be confounding: age, whether sepsis was identified in the ED versus an Inpatient unit, pre-existing CHF, pre-existing CRF, malignancy, lower re- spiratory infection confirmed by chest radiography, altered mentation at presentation, hypoxia, AKI, and coagulopathy. We selected these fac- tors in an effort to control for Comorbidity burden and severity of organ dysfunction at presentation. We also included ED versus inpatient iden- tified sepsis specifically because inpatients with new sepsis would nec- essarily have developed sepsis complicating another medical condition, bundle compliance might be more easily achieved in the ED, and be- cause population-level literature has suggested mortality may be higher for sepsis patients who do not present with sepsis at hospital arrival [26].

We manually entered bundle compliance status into the model, a categorical variable for intermediate vs severe hyperlactemia, and an interaction term between compliance and hyperlactemia, followed by the above-stated covariates. (Non-compliance and intermediate

CT characteristics“>hyperlactemia were set as the reference values, respectively). All terms were manually entered into the model and retained regardless of signif- icance. We assessed goodness-of-fit with Hosmer-Lemeshow test where the null hypothesis, that the model fit the data, was accepted for p N 0.05. We were not able to assess additional potential interactions (e.g., with CHF) without exceeding our allotted degrees of freedom which would have risked over-fitting. We repeated this procedure to test differential risk for 28-day hospital mortality and again for mechan- ical ventilation. There were no missing data for any data-field included in the models.

Sensitivity analysis

As a post-hoc sensitivity analysis, we computed a new compliance variable based solely on the interventional elements of the bundle (i.e., intravenous fluids initiated within 30 min, and antibiotic adminis- tered within both 60 min of time-zero and 180 min of >= 2 SIRS criteria). For this analysis we classified patients as compliant if both antibiotic and fluid initiation goals of the bundle were met, irrespective lactate order-to-result time and blood culture elements. We then ran the

models for 60-day mortality, 28-day mortality, and mechanical ventila- tion as specified above using this alternative compliance definition.

Results

Subject characteristics

The QI database contained records for 7239 patient encounters during the 2015 period. 2417 cases would ultimately be eligible for study analy- ses (Fig. 1). 704(29%) of these patients had a lactate N 3.9 mmol/L. The full compliance rate was 29% within severe and 28% within intermediate hyperlactemia populations. Compliance with fluids and antibiotics bundle elements alone was 47% for severely hyperlactemic patients vs. 36% for the intermediate group. We report distribution of baseline demographic, clinical, and treatment characteristics across the four groups of the cohort in Table 2.

Unadjusted outcomes

Unadjusted outcomes are shown in Table 2. Of the 424(17.5%) in- hospital mortalities occurring within 60 days, we observed 182(14.9%) in the intermediate/non-compliant group versus 41(8.4%) in the inter- mediate/compliant group [difference between compliant and non-com- pliant: 6.5%, confidence interval (CI): 3.1%-9.5%], while the severe/non- compliant group experienced 147(29.2%) versus 54(27.0%) mortalities in the severe/compliant group [difference between compliant and non-compliant: 2.2%, CI:(-)5.4%-9.4%] [difference of differences: 4.3%,

CI: 2.6%-5.9%]. 586(24.2%) total patients required mechanical ventila-

tion: 273(22.3%) intermediate lactate level patients receiving non-com- pliant bundle care versus 63(12.9%) intermediate lactate level/ compliant patients [difference between compliant and non-compliant: 9.4%, CI: 5.4%-13.0%] and 191(37.9%) severe lactate level/non-compli- ant patients versus 59(29.5%) severe lactate level/compliant patients [difference between compliant and non-compliant: 8.4%, CI: 0.6%- 15.7%] [difference of differences: 1.0%, CI: (-)1.6%-3.4% ].

Outcomes on adjustment

Results from adjusted models are summarized in Table 3. As expect- ed, in multiple logistic regression adjusted for: age, whether sepsis was identified in the ED or after admission, CHF, CRF, positive chest radiog- raphy, active malignancy, altered mentation, hypoxia, AKI, and coagu- lopathy, lactate >= 4.0 mmol/L was associated with increased 60-day

mortality (adjusted odds ratio(AOR): 1.99, CI: 1.51-2.61, p b 0.001)

and bundle compliance was associated with decreased 60-day mortality (AOR: 0.62, CI: 0.42-0.90, p = 0.013) (Table 4a). We observed a signif- icant interaction effect between bundle compliance and lactate group, with p_{(interaction)} = 0.022; i.e., the effect of bundle compliance in the model was different for patients with different initial lactate levels. The model displayed adequate fit (Hosmer-Lemeshow: ?²=4.3, p = 0.83). In sensitivity analysis for 60-day hospital mortality where compli- ance was defined only as adherence to the fluids and antibiotics ele- ments of the 3-h bundle, results where similar (Table 4b). Fit

Fig. 1. Flow diagram of patient identification and eligibility.

Table 2

Cohort characteristics and unadjusted outcomes.

Hyperlactemia group Bundle compliance group	(All Subjects) (All subjects)	>= 4.0 mmol/L Non-compliant	>= 4.0 mmol/L Compliant	2.2-3.9 mmol/L Non-compliant	2.2-3.9 mmol/L Compliant
N	2417	504	200	1225	488
Demographics
Male sex	1297 (53.7)	267 (53.0)	115 (57.5)	632 (51.6)	283 (58)
Age - years (IQR)	73 (60, 84)	72 (60,83)	71 (60,82)	74 (61,84)	72.0 (57, 84.5)
BMI - median (IQR)	26.5 (22.4, 31.6)	26.9 (23.1, 31.5)	25.1 (21.9, 29.3)	26.6 (22.4, 32.0)	26.1 (22.2, 31.4)
Medicare	1430 (59.2)	307 (60.9)	106 (53.0)	735 (60.0)	282 (57.8)
Medicaid	180 (7.5)	40 (7.9)	15 (7.5)	77 (6.3)	48 (9.8)
Commercial insurance	782 (32.4)	151 (30.0)	74 (37.0)	405 (33.1)	152 (31.1)
Comorbidities at presentation - no. (%)a
Congestive heart failure	287 (11.9)	76 (15.1)	14 (7.0)	160 (13.1)	37 (7.6)
Chronic obstructive pulmonary disease	183 (7.6)	40 (7.9)	18 (9.0)	95 (7.8)	30 (6.1)
Diabetes	826 (34.2)	176 (34.9)	51 (25.5)	432 (35.3)	167 (34.2)
Immune modifying medications	170 (7.0)	24 (4.8)	14 (7.0)	92 (7.3)	42 (8.6)
Liver failure	50 (2.1)	11 (2.2)	8 (4.0)	22 (1.8)	9 (1.8)
Leukemia, lymphoma, or multiple myeloma	92 (3.8)	18 (3.6)	7 (3.5)	50 (4.1)	17 (3.5)
Metastatic disease	525 (21.7)	160 (31.7)	54 (27.0)	250 (20.4)	61 (12.5)
Organ transplant	17 (0.7)	6 (1.2)	0 (0.0)	8 (0.7)	3 (0.6)
Chronic renal failure	226 (9.4)	48 (9.5)	18 (9.0)	123 (10.0)	37 (7.6)
HIV/AIDS	19 (0.8)	1 (0.2)	0 (0.0)	13 (1.1)	5 (1.0)
Nosocomial infection - no. (%)	113 (4.7)	20 (4.0)	15 (7.5)	58 (4.7)	20 (4.1)
Presentation and acuity - no (%)
Sepsis identified in the emergency department	2046 (84.7)	379 (75.2)	180 (90.0)	1035 (84.5)	452 (92.6)
Suspected site of infection
Urinary	559 (23.1)	96 (19.0)	33 (16.5)	310 (25.3)	120 (24.6)
Respiratory	867 (35.9)	183 (36.3)	69 (34.5)	442 (36.1)	172 (35.5)
Gastrointestinal	258 (10.7)	62 (12.3)	29 (14.5)	131 (10.7)	36 (7.4)
Skin	206 (8.5)	25 (5.0)	11 (5.5)	120 (9.8)	50 (10.2)
Central nervous system	6 (0.2)	2 (0.4)	1 (0.5)	1 (0.1)	2 (0.4)
Other	241 (10.0)	63 (12.5)	28 (14.0)	104 (8.5)	46 (9.4)
Unknown	280 (11.6)	73 (14.5)	29 (14.5)	117 (9.6)	61 (12.5)
Positive chest radiography at presentation	788 (33.3)	151 (30.0)	68 (34.0)	400 (32.7)	169 (34.6)
Initial lactate (mmol/L) - median (IQR)	3.1 (2.6, 4.3)	5.3 (4.6, 6.85)	5.2 (4.4, 6.8)	2.8 (2.4, 3.2)	2.8 (2.4, 3.2)
Acute kidney injuryb	327 (13.5)	90 (17.9)	38 (19.0)	145 (11.8)	54 (11.1)
Coagulopathyc	311 (12.9)	76 (15.1)	32 (16.0)	149 (12.2)	54 (11.1)
Bilirubin N 2.0 mg/dL	169 (7.0)	48 (9.5)	12 (6.0)	77 (6.2)	33 (6.8)
Altered mental status	393 (16.3)	106 (20.1)	40 (20.0)	192 (15.7)	55 (11.3)
Hypoxiad	315 (13.0)	103 (20.4)	28 (14.0)	156 (12.7)	28 (5.7)
Super SIRS criteria at triage - no. (%)?	538 (22.4)	126 (25.4)	46 (23.0)	278 (22.8)	88 (18.0)
Interventionse Time to fluids initiation - medianf (IQR)	7 (-44, 65)	35.5 (-52, 118)	-22 (-48.5, 5)	44 (-23, 150)	-23.5 (-53.5, 5)
Fluid initiation <= 30 min	1275 (52.8)	190 (37.7)	200 (100)	397 (32.4)	488 (100)
Fluid volume given (Liters) - median (IQR)	1.5 (1.0; 2.0)	1.5 (0.5, 2.5)	2.0 (1.5, 3.0)	1.0 (0.5, 2.0)	2.0 (1.0, 2.125)
Lactate order to result time - median (IQR)	44 (21,86)	53 (21, 122.5)	34.5 (21, 51)	49 (21, 108)	37 (21, 55)
Lactate order to result <= 90 min	1862 (77.0)	329 (65.3)	200 (100)	845 (69.0)	488 (100)
Blood cultures drawn before antibiotics	1662 (68.8)	282 (56.0)	200 (100)	692 (56.5)	488 (100)
time to antibiotic administration - median (IQR)	52 (-2, 137)	70 (11, 229)	25 (-6.5, 59.5)	70 (1, 204.5)	22 (-11, 65.5)
Antibiotics <= 60 min (time-zero) and <= 180 min(2 SIRS)	1805 (74.7)	325 (64.5)	200 (100)	792 (64.7)	488 (100)
Both fluids & antibiotic bundle goal compliance	1094 (45.3)	134 (26.6)	200 (100)	272 (22.2)	488 (100)
No. of bundle elements accomplished - no. (%)
Zero	73 (3.0)	19 (3.8)	0 (0)	54 (4.4)	0 (0)
One	323 (13.4)	86 (17.1)	0 (0)	237 (19.3)	0 (0)
Two	470 (19.4)	157 (31.2)	0 (0)	313 (25.6)	0 (0)
Three	863 (35.7)	242 (48.0)	0 (0)	621 (50.7)	0 (0)
All four	688 (28.5)	0 (0)	200 (100)	0 (0)	488 (100)
Unadjusted Outcomes
60-day hospital mortality - no. (%)	424 (17.5)	147 (29.2)	54 (27.0)	182 (14.9)	41 (8.4)
28-day hospital mortality - no. (%)	388 (16.1)	138 (27.4)	52 (26.0)	162 (13.2)	36 (7.4)
Mechanical ventilation required - no. (%)	586 (24.2)	191 (37.9)	59 (29.5)	273 (22.3)	63 (12.9)
Vasopressors administered - no. (%)	262 (11.1)	111 (22.0)	20 (10.0)	112 (9.1)	19 (3.9)
Median length of stay - days (IQR)	7 (4, 13)	7.5 (4,14)	7 (3,14)	7 (4,12)	6 (4,11)

^a Comorbidities reflect status at time-zero, and would not reflect conditions that developed subsequently during hospital stay.

^b Acute kidney injury defined as creatinine N 2.0 or 50% increase from a known baseline.

^c Coagulopathy defined as platelet count b 150 000 cells/um³, international normalized ratio N 1.5, or partial thromboplastin time N 60 s.

^d Hypoxia defined as PaO₂/FiO₂ b 300 or an increased O₂ requirement to maintain SaO₂ N 90%.

^e All times are in minutes, and reflect the time elapsed from time-zero unless otherwise indicated. A negative time indicates an intervention performed before time-zero. E.g., a patient whose fluid resuscitation began 20 min before laboratory results indicating organ dysfunction became available have a fluid initiation time of -20.

^f All crystalloid was 0.9% normal saline solution.

^? Super-SIRS” defined as >=2 of the following criteria at triage: heart rate >=120, respiratory rate >=24, temperature N 38.3 (Celsius) or b36.0 (Celsius), and acutely altered mental status.

Table 3

Adjusted outcomes from multivariable logistic regression analyses

Table 4b

Final sensitivity model for predictors of 60-day in-hospital mortality.

Outcome	Model fit	Factor tested	Odds ratio	95% CI	p Value		Variable	Odds ratio	95% Confidence-interval		p Value
Primary outcome						Lactate N 3.9 mmol/L		1.72	1.25	2.37	0.001
60-day	?2=4.3,	Bundle compliance	0.62	0.42-0.90	0.013	Compliance with antibiotic and fluids		0.70	0.51	0.96	0.029
in-hospital	p = 0.83					bundle elements
mortality		Lactate >= 4.0 mmol/L	1.99	1.51-2.63	b0.001	Antibiotic and fluids		2.02	1.24	3.28	0.004

compliance x Lactate Interaction

Age (per 10 years)	1.27	1.17	1.38	b0.001
Initial sepsis episode not in the ED	3.70	2.70	5.00	b0.001
Congestive heart failure	1.76	1.37	2.26	b0.001
Chronic renal failure	1.16	0.83	1.60	0.39
Active malignancy	1.24	0.86	1.79	0.25
Positive chest radiography	1.83	1.06	3.16	0.030
Acutely altered mental status	1.40	1.05	1.88	0.022
Hypoxiaa	1.47	1.07	2.01	0.016
Acute kidney injuryb	1.90	1.40	2.59	b0.001
Coagulopathyc	2.14	1.58	2.90	b0.001

Compliance * Lactate Interaction	1.91	1.10-3.31	0.022
econdary outcomes 8-day ?2=9.4, Bundle compliance 0.59 0.39-0.88 0.010 in-hospital p = 0.31

S

2

mortality		Lactate >= 4.0 mmol/L	2.14	1.61-2.84	b0.001
		Compliance * Lactate	1.99	1.13-3.50	0.017
Mechanical	?2=11.4,	Interaction Bundle compliance	0.69	0.50-0.95	0.024

ventilation

p = 0.20

Lactate >= 4.0 mmol/L 1.69 1.30-2.22 b0.001

Sensitivity analyses

Compliance * Lactate Interaction

1.29 0.77-2.17 0.34

Hosmer and Lemeshow test for goodness-of-fit: ?²=11.6, p = 0.20.

Age was a continuous variable with a base unit equal to 10 years. All other variables were treated as binary variables with the negative state set as the referent. There were 424 events occurring in the sample.

60-day

in-hospital mortality

?²=11.6,

p = 0.20

Antibiotic and fluids 0.70 0.51-0.96 0.029

compliance

Lactate >= 4.0 mmol/L 1.72 1.25-2.37 0.001

^a Hypoxia defined as PaO₂/FiO₂ b 300 or an increased O₂ requirement to maintain SaO₂ N 90%.

^b Acute Kidney Injury defined as creatinine N 2.0 or 50% increase from a known baseline.

28-day

in-hospital

?²=7.4, p = 0.49

Compliance * Lactate Interaction Antibiotic and fluids

compliance

2.02 1.24-3.28 0.004

0.65 0.47-0.91 0.013

^c Coagulopathy defined as platelet count b 150 000 cells/um³ or an international nor- malized ratio N 1.5 or a partial thromboplastin time N 60 s.

mortality

Lactate >= 4.0 mmol/L 1.86 1.34-2.58 b0.001

remained adequate (Hosmer-Lemeshow: ?²=11.6, p = 0.20), and we

Mechanical ventilation

?²=18.1, p = 0.02

Compliance * Lactate Interaction Antibiotic and fluids

compliance

2.07 1.26-3.40 0.004

0.83 0.63-1.10 0.189

observed significant association between the outcome and lactate group (AOR: 1.72, CI: 1.25-3.72, p = 0.001), antibiotic and fluids com- pliance (AOR: 0.70, CI: 0.51-0.96, p = 0.029), and the interaction of lac-

Lactate >= 4.0 mmol/L 1.66 1.22-2.25 b0.001

Compliance * Lactate 1.20 0.76-1.90 0.43 Interaction

Models were adjusted for the following variables: age, whether sepsis was identified in the emergency department, pre-existing heart failure, pre-existing renal failure, active malignan- cy, whether there was a positive chest radiographic finding at the time of initial sepsis epi- sode, acutely altered mental status, hypoxia, acute kidney injury, and coagulopathy. Hosmer-Lemeshow tests assessed goodness-of-fit, where the null hypothesis that the model fit the data was accepted for p N 0.05.

The interaction coefficient is the ratio of the odds ratio for a variable at one value of a second variable over the odds ratio for the first variable at the alternate value of the second variable. A significant, positive number in this table therefore indicates that the odds ratio for bundle compliance increases from the reference value of lactate (2.2-3.9) to a larger number for the alternate value of lactate (>=4.0), i.e., a less pronounced association.

Table 4a

Final logistic model for predictors of 60-day in-hospital mortality.

tate group with antibiotic and fluids compliance (p_{(interaction)} = 0.004). Results were similar when 28-day hospital mortality was the out- come (Table 3, Tables 1aS, 1bS). In a model testing predictors of me- chanical ventilation (Hosmer-Lemeshow: ? ²=11.4, p = 0.20), both compliance and initial lactate were significantly associated with the outcome (AOR: 0.69, CI: 0.50-0.95, p = 0.024; and AOR: 1.69, CI:

1.30-2.20, p b 0.001, respectively), but the interaction between the two variables was not significant on adjustment (p_{(interaction)} = 0.34). In sensitivity analysis, the model assessing mechanical ventilation did not adequately fit the data (Hosmer-Lemeshow: ?²=18.1, p = 0.02) (Table 3, Tables 2aS, 2bS).

Discussion

Variable

Lactate N 3.9 mmol/L	1.99	1.51	2.63	b0.001
Full 3-h bundle compliance	0.62	0.42	0.90	0.013
Bundle compliance x Lactate	1.91	1.10	3.31	0.022
Interaction
Age (per 10 years)	1.27	1.17	1.37	b0.001
Initial sepsis episode not in the ED	3.57	2.63	4.76	b0.001
Congestive heart failure	1.14	0.82	1.58	0.44
Chronic renal failure	1.23	0.86	1.78	0.26
Active malignancy	1.80	1.04	3.12	0.035
Positive chest radiography	1.79	1.40	2.29	b0.001
Acutely altered mental status	1.39	1.04	1.86	0.025
Hypoxiaa	1.45	1.06	1.98	0.020
Acute kidney injuryb	1.93	1.41	2.62	b0.001
Coagulopathyc	2.14	1.58	2.90	b0.001

Odds ratio

95%

Confidence-interval

p

Value

In this multisite analysis of 2417 non-hypotensive sepsis patients, the effect of initial bundle compliance was associated with a larger mor- tality risk-reduction for patients with intermediate hyperlactemia than for patients with lactate N 3.9 mmol/L. However, the magnitude of re- duced mechanical ventilation risk associated with bundle compliance was comparable between lactate groups. We cautiously interpret this to suggest that the population with the less severe presentation derived greater benefit from early application of initial bundle care.

Importantly, this analysis excluded patients who were hypotensive

at their initial sepsis presentation. We made this exclusion due to the observational nature of the study. In the context of the study question, including these patients would present substantial issues of data endogeneity. Clinically, hypotensive patients are more easily identified

Hosmer and Lemeshow test for goodness-of-fit: ? ²=5.3, p = 0.73. Age was a continuous variable with a base unit equal to 10 years. All other variables were treat- ed as binary variables with the negative state set as the referent. There were 424 events occur- ring in the sample.

^a Acute kidney injury defined as creatinine N 2.0 or 50% increase from a known baseline.

^b Coagulopathy defined as platelet count b 150 000 cells/um³ or an international nor- malized ratio N 1.5 or a partial thromboplastin time N 60 s.

^c Hypoxia defined as PaO₂/FiO₂ b 300 or an increased O₂ requirement to maintain SaO₂ N 90%.

as septic in their initial presentation, suggesting hypotension would have been associated with bundle compliance [27]. Hypotension is often concomitant with, and many times a likely cause of, tissue hypo- perfusion and contemporary thinking suggests hypotension resulting from systemic inflammatory vasodilation and microvascular thrombo- sis to be an important mechanism of injury in septic shock [28]. As a re- sult, we would have expected a second confounding association, between hypotension and more severe hyperlactemia. Further,

hypotension is a predictor of mortality [12,13,21], suggesting at least three major dimensions of bias had these patients been included.

This exclusion also presents a key lens for interpreting our findings. For one, this is likely a factor explaining the low mortality rate for the in- termediate hyperlactemia group. More interestingly, we observe high mortality in the severe hyperlactemia group, supporting prior literature that this non-hypotensive cryptic shock population is at high risk for mortality. However, the significant interaction-coefficient between lac- tate and bundle compliance indicated the association of reduced mor- tality and bundle compliance was significant only for patients with intermediate hyperlactemia in this study. While it is possible this may be explained by the smaller sample size for the severe hyperlactemia group, or by unmeasured differences between groups, the severe hyperlactemia patients may have simply not responded as well to Initial resuscitation. Should this observation prove reproducible, this would be new evidence supporting the framework that a less severe presentation of sepsis benefits from equal rates of successful adherence to a 3-h sep- sis bundle.

We also note that a full 50% of the cohort assessed were intermediate hyperlactemia patients receiving non-compliant care. While likely in- flated by the exclusion of initially hypotensive patients, this large pro- portion nevertheless represents a large target population for future quality initiatives, regardless of whether Treatment response truly dif- fers by initial hyperlactemia severity. Importantly, the low overall com- pliance rates, particularly given that the study population is drawn from high-performing sites [8,22,24], suggest hemodynamically stable sepsis patients may be at heightened risk of under-recognition and under- treatment in general. Notably, our 29% compliance rates are comparable to the pre-implementation rates in a prior large study by Liu and col- leagues that assessed an initiative specifically targeting intermediate lactate patients. Our 44% adherence to antibiotics and fluids in the inter- mediate lactate cohort is comparable to that study’s post-implementa- tion adherence rates.

Our results suggest that early intervention for less severely ill pa-

tients could put them in better position with respect to disease progres- sion, supporting the approach to sepsis care that emphasizes prevention of delayed decompensation. Our results are consistent with the paper by Liu and colleagues that suggested hemodynamically stable sepsis pa- tients with intermediate lactate may derive benefit from initial bundle care [20], although we are unaware of a direct comparison of this pop-

ulation to a “cryptic shock” population with lactate >= 4.0. These findings warrant further exploration. Future studies should attempt to deter-

mine the differential effect of bundle compliance in a larger population using more generalizable definitions of sepsis and more comprehensive characterizations of patients’ physiologic status at the time of presentation.

Interestingly, the interaction effect was significant only for mortality, while the association between bundle compliance and reduced risk for mechanical ventilation was preserved across lactate groups. We could not investigate this seemingly contradictory finding further because we were unable to assess other organ-support requirements as out- comes, such as renal replacement therapy. It could be non-respiratory organ dysfunction contributes disproportionately to mortality in sepsis, and although we are not in a position to evaluate this hypothesis with our data, future studies should consider this question as well.

This investigation has a number of limitations. As with all non-ran- domized studies, unintended bias presents an important concern. We control for observed factors with multivariable modeling, but potential for unobserved confounding remains. No physician intends to adminis- ter delayed care, so patients receiving compliant versus non-compliant care may be somehow “different”, with the latter perhaps demonstrat- ing a more complex or insidious presentation. We also employ inclusion criteria that align with the direction of Sepsis-3 definitions but that are not entirely concordant with either new or old consensus criteria, limit- ing direct integration of our findings. We were not able to use these criteria for operational reasons. Also, our window of observation focuses

on the initial sepsis episode and care, but downstream outcomes like mortality can be influenced by factors occurring later in the patient’s stay. In the context of this study’s research question, we also acknowl- edge the inability to capture composite-measures such as SOFA scores a secondary limitation that probably does not intrinsically limit inter- nal-validity, but that poses an obstacle in assessing external-validity. Additional limitations include retrospective design, inability to assess other measures of organ support besides mechanical ventilation, and in- ability to assess inter-rater reliability among QI database abstractors.

In conclusion, bundle compliance may be low for non-hypotensive sepsis patients with elevated lactate, suggesting this population could benefit from targeted quality initiatives. We observed a greater associa- tion between bundle compliance and improved mortality for patients with intermediate compared to severe elevations in initial lactate level. Our results support rapid sepsis bundle application for stable, in- fected patients with intermediate hyperlactemia.

Funding

This investigation was funded in part by a grant from the Center for Medicare and Medicaid Innovation to the High Value Healthcare Collab- orative, of which the study sites’ umbrella health system was a part. This grant helped fund the underlying quality improvement program and database in this study.

Conflict of interest

The authors have no financial conflicts of interest to disclose.

Appendix A. Supplementary data

Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.ajem.2017.01.029.

References

1. Fleischmann C, Scherag A, Adhikari NK, et al. Assessment of global incidence and mortality of hospital-treated sepsis. Current estimates and limitations. Am J Respir Crit Care Med 2016;193(3):259-72.
2. Kumar A, Roberts D, Wood KE, et al. Duration of hypotension before initiation of ef- fective antimicrobial therapy is the critical determinant of survival in human septic shock. Crit Care Med 2006;34(6):1589-96.
3. Jones AE, Brown MD, Trzeciak S, et al. The effect of a quantitative resuscitation strat- egy on mortality in patients with sepsis: a meta-analysis. Crit Care Med 2008; 36(10):2734-9.
4. Ferrer R, Artigas A, Suarez D, et al. Effectiveness of treatments for severe sepsis: a prospective, multicenter, observational study. Am J Respir Crit Care Med 2009; 180(9):861-6.
5. Ferrer R, Martin-Loeches I, Phillips G, et al. Empiric antibiotic treatment reduces mortality in severe sepsis and septic shock from the first hour: results from a guide- line-based performance improvement program. Crit Care Med 2014;42(8):1749-55.
6. Seymour CW, Cooke CR, Heckbert SR, et al. Prehospital intravenous access and fluid resuscitation in severe sepsis: an observational cohort study. Crit Care 2014;18(5): 533.
7. Rhodes A, Phillips G, Beale R, et al. The surviving sepsis campaign bundles and out- come: results from the international multicentre prevalence study on sepsis (the IMPreSS study). Intensive Care Med 2015;41(9):1620-8.
8. Leisman D, Wie B, Doerfler M, et al. Association of fluid resuscitation initiation with- in 30 minutes of severe sepsis and septic shock recognition with reduced mortality and length of stay. Ann Emerg Med 2016.
9. Loiacono LA, Shapiro DS. Detection of hypoxia at the cellular level. Crit Care Clin 2010;26(2):409-21 (table of contents).
10. Howell MD, Donnino M, Clardy P, Talmor D, Shapiro NI. Occult hypoperfusion and mortality in patients with suspected infection. Intensive Care Med 2007;33(11): 1892-9.
11. Mikkelsen ME, Miltiades AN, Gaieski DF, et al. Serum lactate is associated with mor- tality in severe sepsis independent of organ failure and shock. Crit Care Med 2009; 37(5):1670-7.
12. Seymour CW, Liu VX, Iwashyna TJ, et al. Assessment of clinical criteria for sepsis: for the third international consensus definitions for sepsis and septic shock (Sepsis-3). JAMA 2016;315(8):762-74.
13. Shankar-Hari M, Phillips GS, Levy ML, et al. Developing a new definition and assessing new clinical criteria for septic shock: for the third international consensus definitions for sepsis and septic shock (Sepsis-3). JAMA 2016;315(8):775-87.
14. Jones AE, Shapiro NI, Trzeciak S, et al. Lactate clearance vs central venous oxygen sat- uration as goals of early sepsis therapy: a randomized clinical trial. JAMA 2010; 303(8):739-46.
15. Nguyen HB, Rivers EP, Knoblich BP, et al. Early lactate clearance is associated with improved outcome in severe sepsis and septic shock. Crit Care Med 2004;32(8): 1637-42.
16. Bone RC, Balk RA, Cerra FB, et al. Definitions for sepsis and organ failure and guide- lines for the use of innovative therapies in sepsis. The ACCP/SCCM consensus confer- ence committee. American College of Chest Physicians/Society of Critical Care Medicine. Chest 1992;101(6):1644-55.
17. Levy MM, Fink MP, Marshall JC, et al. 2001 SCCM/ESICM/ACCP/ATS/SIS International Sepsis Definitions Conference. Crit Care Med 2003;31(4):1250-6.
18. Jones AE, Puskarich MA. The surviving sepsis campaign guidelines 2012: update for emergency physicians. Ann Emerg Med 2014;63(1):35-47.
19. National Quality Forum. Severe Sepsis and Septic Shock: Management Bundles (NQF

    #0500).

    Liu VX, Morehouse JW, Marelich GP, et al. Multicenter implementation of a treat- ment bundle for sepsis patients with intermediate lactate values. Am J Respir Crit Care Med 2015.

20. Singer M, Deutschman CS, Seymour CW, et al. The third international consensus def- initions for sepsis and septic shock (Sepsis-3). JAMA 2016;315(8):801-10.
21. Doerfler ME, D’Angelo J, Jacobsen D, et al. Methods for reducing sepsis mortality in emergency departments and inpatient units. Jt Comm J Qual Patient Saf 2015; 41(5):205-11.
22. Statewide Planning and Research Cooperative System (SPARCS). In: Health NYSDo2016.
23. Leisman DE, Doerfler ME, Ward MF, et al. Survival benefit and cost savings from compliance with a simplified 3-hour sepsis bundle in a series of prospective, multi- site, observational cohorts. Crit Care Med 2016.
24. Mebazaa A, Laterre PF, Russell JA, et al. Designing phase 3 sepsis trials: application of learned experiences from critical care trials in acute heart failure. J Intensive Care 2016;4:24.
25. Liu V, Escobar GJ, Greene JD, et al. Hospital deaths in patients with sepsis from 2 in- dependent cohorts. JAMA 2014;312(1):90-2.
26. Amaral AC, Fowler RA, Pinto R, et al. Patient and organizational factors associated with delays in antimicrobial therapy for septic shock. Crit Care Med 2016.
27. Angus DC, van der Poll T. Severe sepsis and septic shock. N Engl J Med 2013;369(9): 840-51.