Journal Logo

Feature Articles

The Association Between Antibiotic Delay Intervals and Hospital Mortality Among Patients Treated in the Emergency Department for Suspected Sepsis*

Taylor, Stephanie Parks MD1; Anderson, William E. MS2; Beam, Kent MD1; Taylor, Brice MD1; Ellerman, Justin MD3; Kowalkowski, Marc A. PhD2

Author Information
doi: 10.1097/CCM.0000000000004863


Prompt administration of antibiotics is a cornerstone of therapy for patients with suspected sepsis (1). Retrospective and prospective data suggest that every 1-hour delay in antibiotics after emergency department (ED) triage or the onset of organ dysfunction or shock may lead to a 3–7% increase in the odds of a poor outcome (2–4). Prompted by these data, a 2018 update of the Surviving Sepsis Campaign (SSC) guidelines recommended antibiotic initiation within 1 hour of ED triage (1). The Centers for Medicare and Medicaid Services Early Management Bundle for Severe Sepsis/Septic Shock (SEP-1) core quality measure specifies that antibiotics be received within 3 hours of identification of sepsis, which is defined by a complex set of organ dysfunction criteria (5).

Although timely administration of antibiotics for patients with suspected sepsis is a sensible strategy, it has been argued that 1 hour is neither a feasible nor appropriate target for all patients (6,7). Further, it is unclear whether ED triage to antibiotic administration is the best time interval to measure and target (6). An alternative proposition expressed by the Infectious Disease Society of America (IDSA) is “to minimize the time interval between each stat antibiotic order for septic shock and initiation of the infusion” (8). Indeed, some data suggest delays from antibiotic order to infusion are common and associated with increased mortality (9). However, delay in recognition of sepsis, represented by the time from ED triage to antibiotic order, is the largest contributor to the total interval from ED triage to antibiotic infusion (10). Because delays in both clinician recognition of sepsis and in prompt delivery of antibiotics contribute to total delay, we aimed to determine the contribution of each to mortality.


Study Design

We performed a retrospective study using data from 12 hospitals in a large southeastern U.S. healthcare system. During the study period, a multidimensional sepsis identification and treatment program matured at our institution, including provider education, sepsis-specific order sets, quality assurance review, and deployment of Sepsis Coordinators at some sites. The institutional review board of the Carolinas Medical Center approved the study design (Institutional Review Board number 08-17-03E).

Patient Selection

We selected adult (≥ 18 yr old) patients who presented to the ED and were hospitalized with clinically suspected sepsis between January 2014 and September 2017. We defined “suspected sepsis” as both: 1) digital signature of infection and 2) digital signature of organ dysfunction (11). For infection, we specified: 1) antibiotics ordered and administered within 12 hours of ED triage and 2) culture ordered within 48 hours of ED triage. We then applied the CDC Adult Sepsis Event Sequential Organ Failure Assessment Criteria (eSOFA) criteria to capture organ dysfunction (12), defined as one or more eSOFA criteria present within 6 hours (without reference to baseline values). We aimed to specify eligibility determination that approximated the alignment of eligibility, exposure, and start of follow-up times (13).

The primary outcome was hospital mortality, which included discharge to hospice. The primary exposures were as follows: 1) recognition time (time in hours from ED triage to antibiotic order) and 2) administration time (time in hours from antibiotic order to antibiotic infusion). Antibiotic order and administration times were obtained from the electronic health record.

We included a priori identified covariates as potential confounders between task delay and outcome based on prior studies. These were age; race; Charlson comorbidity index; infection site (determined by diagnosis codes and categorized as genitourinary, abdominal, respiratory, bloodstream, skin and soft tissue, or other); temperature (hypothermic, normothermic, hyperthermic, or not available); minimum mean arterial pressure (MAP), maximum respiratory rate; platelet count; serum lactate (< 2, 2–4, > 4 mmol/L, and not available); and intubation. We included hospital as a random effect to account for clustering of effects within hospitals. We limited the ascertainment time to within 6 hours from ED presentation to avoid adjusting for variables that were mediators of the effect of the exposures (antibiotic delay intervals) on the outcome.

Missing Data

We had complete data for most variables. For variables with less than or equal to 1% missing data (MAP and platelet count), we imputed the median value from the group characterized by the patient’s sex, race, Charlson Comorbidity Index, and body mass index. To account for informative missingness associated with greater than 1% missing values (serum lactate and temperature), we converted the variable to categorical format and classified missing values as “not available”.

Statistical Analysis

We summarized demographic and clinical characteristics of the sample using means (sd) for normally distributed variables, medians (interquartile range [IQR]) for continuous variables found not to be normally distributed, and counts (percentage) for categorical variables. We modeled the association between time intervals and hospital mortality using generalized linear mixed models (GLMMs) to adjust for potential correlation among observations within the same hospital and among multiple encounters of an individual patient. As the dependent variable in the models was binary, we used a logit link function and variance function for the binomial distribution. We assumed an exchangeable working correlation structure. The models included as fixed effects the time interval of interest plus the potential confounding variables specified above, except that we transformed minimum MAP, Charlson comorbidity index, and platelet count to their natural logarithms because the Box-Tidwell test revealed nonlinear relationships between these variables and the logit. In collinearity diagnostics, we did not observe linear associations among the independent variables. We chose “less than one hour” as the reference for recognition time to align with SSC recommendations and “less than 0.5 hours” as the reference for administration time based on clinical judgment and inspection of the variable distribution. The GLMMs generated odds ratios (ORs) corresponding to the pairwise comparisons, as well as CIs adjusted for multiple comparisons using Dunnett’s method (14). We performed subgroup analyses repeating the same methods above: 1) for the subset of patients with a sepsis diagnosis code at discharge and 2) for patients meeting suspected septic shock criteria (i.e., MAP < 65 and serum lactate level > 2 mmol/L). We conducted all analyses using SAS Enterprise Guide 7.1 (SAS Institute, Cary, NC). All tests of significance were two-sided, and a p value of less than 0.05 was considered statistically significant.


Characteristics of Patients

The cohort included 28,865 unique encounters representing 24,093 patients presenting the ED with suspected sepsis. Table 1 shows descriptive results to illustrate the cohort. Table 2 shows values for the primary exposure variables. The median time from ED arrival to antibiotic administration (total delay) was 3.4 hours (IQR, 2.0–6.0 hr). This delay comprised a median recognition delay (ED triage to order time) of 2.7 hours (1.5–4.7 hr) and median administration delay (antibiotic order to infusion time) of 0.6 hours (0.3–1.2 hr).

TABLE 1. - Characteristics of the Cohort of Patients With Suspected Sepsis Presenting to the Emergency Department (n = 24,093 Unique Patients, Most Recent Encounter)
Variables Values
Age, yr, mean (sd) 59.7 (17.3)
Female, n (%) 12,791 (53.1)
Race, n (%)
 African American 5,879 (24.4)
 Asian 194 (0.8)
 Caucasian 16,720 (69.4)
 Other 1,122 (4.7)
 Unknown 178 (0.7)
Weighted Charlson sum, mean (sd) 5.0 (3.8)
Maximum respiratory rate, median (IQR) 22.0 (20.0–28.0)
Minimum mean arterial pressure, mm Hg,a median (IQR) 73.3 (64.0–85.0)
Temperature, °F, n (%)
 Hypothermic (< 96.8) 1,325 (5.5)
 Normothermic (96.8–100.4) 16,553 (68.7)
 Hyperthermic (> 100.4) 5,738 (23.8)
 Not available 477 (2.0)
Platelet count, median (IQR) 229.0 (170.0–298.0)
Maximum lactate, mmol/La, n (%)
 < 2 8,006 (33.2)
 2–4 6,401 (26.6)
 > 4 2,760 (11.5)
 Not available 6,926 (28.8)
Intubated, n (%) 1,922 (8.0)
Hospital mortality, n (%)b 2,378 (9.9)
 In-hospital death 1,565 (6.5)
 Discharged to hospice 813 (3.4)
IQR = interquartile range.
aWithin 6 hr of emergency department arrival.
bIncluding discharge to hospice.
Missing data: mean arterial pressure 0.4%, platelet count 3.8%.

TABLE 2. - Length of Antibiotic Delay Intervals Among Patients With Suspected Sepsis Presenting to the Emergency Department
Intervals All Encounters (in = 28,865) Septic Shock Encounters (n = 3,681)
Triage to administration time (total delay), hr, median (IQR) 3.4 (2.0–6.0) 2.3 (1.4–4.1)
 Triage to order time (recognition delay), hr 2.7 (1.5–4.7) 1.7 (0.9–3.1)
 Order to infusion time (administration delay), hr 0.6 (0.3–1.2) 0.5 (0.2–1.0)
IQR = interquartile range.

Figure 1 shows results from the GLMMs. Recognition delay was significantly associated with hospital mortality (< 0.01), with pairwise comparisons revealing nonsignificant effects for recognition times of 1–3 hours and 3–6 hours but an increased odds of hospital mortality for a recognition time greater than 6 hours, compared with less than 1 hour (OR, 1.21; 95% CI, 1.0–1.46). Administration delay was also significantly associated with hospital mortality (p < 0.01), and pairwise comparisons revealed significantly increased odds of hospital mortality only for order to infusion times of 1.5–2 hours (OR, 1.35; 95% CI, 1.07–1.71) and 2–2.5 hours (OR, 1.51; 95% CI, 1.14—2.00) compared with less than 0.5 hours.

Figure 1.
Figure 1.:
Association between discrete intervals of delay and hospital mortality in patients presenting to the emergency department (ED) with suspected sepsis. Recognition delays are compared with reference less than 1 hr, and administration delays are compared with reference less than 0.5 hr. OR = odds ratio.

In 10,548 encounters (37%), patients had a discharge diagnosis of sepsis, and in 3,681 encounters (15.3%), patients met the criteria for suspected septic shock. Applying the same methods to these subgroups revealed similar findings in the encounters involving a discharge diagnosis of sepsis as the full cohort (Supplemental Table 1, Supplemental Digital Content 1, For the group with septic shock, we found no association between total delay and mortality (p = 0.21) or recognition delay and mortality (p = 0.17), but administration delay was associated with mortality (p = 0.02) (Supplemental Table 2, Supplemental Digital Content 1,


A key component of guideline-recommended care in sepsis is the prompt administration of appropriate antibiotics (1). However, there is no consensus for which time intervals should be measured and minimized. In order to better understand the relationship between time to antibiotics and outcomes as well as unpack the complexities of the task of antibiotic administration, we considered two intervals: ED arrival to antibiotic order placement (recognition delay) and order placement to antibiotic infusion (administration delay).

Our study provides insights regarding which time to antibiotic intervals should be targeted for patients with suspected sepsis. Experts have suggested focusing on minimizing administration delay (time from antibiotic order to infusion) because this interval is associated with mortality (9) involves a concrete “time zero” and is a straightforward process measure that hospitals can work to improve. Indeed, we found an association between administration delay and mortality for patients with suspected sepsis in this study. However, administration delay comprised only one fifth of the total delay, whereas recognition delay (time from arrival to antibiotic order) comprised four fifths of total delay. Recognition delay was also associated with mortality, with recognition delay greater than 6 hours associated with 21% increased odds of mortality compared with recognition time less than 1 hour. Thus, sole focus on administration time as a quality measure for patients with suspected sepsis may miss opportunities to improve sepsis care. Notably, for the subgroup of patients with suspected septic shock, administration delay was the only interval found to have a statistically significant association with mortality. This may be due to more uniform early recognition and readiness to order antibiotics in patients who are more obviously severely ill.

Importantly, our effect estimates and measurements of precision plausibly exclude meaningful harms from delays less than 6 hours compared with 1 hour, suggesting that a 1-hour target from ED triage for patients with suspected sepsis may be too aggressive. This supports the SSC modification of their recommendations to advocate delivery of antibiotics within 1 hour of sepsis recognition, rather than ED triage (15). Because nearly 40% of patients initially treated for sepsis may ultimately have an alternative condition (16,17), our results suggest that allowing longer time for additional diagnostic investigations prior to antibiotic use may be reasonable for patients with suspected sepsis (7). Further, our results support the recent IDSA position that reflexive antibiotic administration for all patients with possible sepsis risks increasing excessive antibiotic utilization without clear benefit (8).

Our findings differ from prior studies showing an association between time to antibiotics and mortality in sepsis. This could be due to differences in case mix between our cohort and others, as the benefit of early antibiotics likely varies based on severity of illness. However, we found similar results for subgroups including patients with a discharge diagnosis of sepsis and those with septic shock. Additionally, our findings could represent a ceiling effect in a healthcare system with an already high-functioning sepsis treatment pathway. Although these factors may be contributing to the lack of association with mortality for shorter antibiotic delays, an important distinction between this analysis and those in prior studies is that we provide distinct adjusted estimates for the association between time intervals and mortality rather than using blended estimates across long time intervals. Using these blended estimates to conclude that “every hour counts” can be misleading because models average the minimal or absent hour-by-hour mortality rates in the first few hours of delay with large increases in mortality seen with very long delays (18).

Our study has several strengths. We deliberately selected our “suspected sepsis” cohort to emulate the population for which initial antibiotic treatment recommendations are being applied. This is critical in capturing the true effect of antibiotic administration in actual practice. Studies that include only patients ultimately deemed to have sepsis (e.g., using diagnosis codes or requiring multiple days of antibiotic administration) ignore the absent or harmful effect of antibiotics accrued to patients with suspected sepsis but ultimately diagnosed with something else. We evaluated discrete rather than linearized associations between time to antibiotics and mortality intervals to avoid artificial hourly associations. We purposefully selected potential confounders that plausibly 1) were associated with the exposure of interest, 2) caused or contributed to the outcome of interest, and 3) did not reside in the causal pathway between the exposure and outcome and avoided adjusting for variables that occurred downstream to antibiotic exposure that might represent mediators rather than confounders.

Our study has notable limitations. First, the observed associations may be due to residual confounding as important confounders may not be fully accounted for in these observational data. Second, although we purposefully selected potential confounders, there are many complex and uncertain factors in the pathway between antibiotic delays and mortality, which make causal inference challenging. Third, our analysis uses data from 12 diverse hospitals but within a single healthcare system; thus, organizational factors common to the system may limit generalizability. Fourth, we evaluated patients with suspected sepsis based on broad clinical criteria, and our analysis of patients with septic shock was a secondary analysis with a smaller cohort. Thus, readers should use caution when applying these to patients with septic shock, in whom the consequences of antibiotic delay are likely much more severe.


Delays in both recognition time and administration time contribute to the delay in antibiotic delivery for patients with suspected sepsis. In this study, delays in both intervals were associated with increased hospital mortality, but the association was only for longer delays. These results suggest that both metrics may be important to measure and improve but do not support a 1-hour antibiotic administration target for patients with suspected sepsis.


1. Levy MM, Evans LE, Rhodes A. The surviving sepsis campaign bundle: 2018 update. Crit Care Med. 2018; 46:997–1000
2. 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 guideline-based performance improvement program. Crit Care Med. 2014; 42:1749–1755
3. Kumar A, Roberts D, Wood KE, et al. Duration of hypotension before initiation of effective antimicrobial therapy is the critical determinant of survival in human septic shock. Crit Care Med. 2006; 34:1589–1596
4. Seymour CW, Gesten F, Prescott HC, et al. Time to treatment and mortality during mandated emergency care for sepsis. N Engl J Med. 2017; 376:2235–2244
5. Centers for Medicare & Medicaid Services. CMS Specifications Manual Version 5.2a. 2020. Available at: Accessed November 15, 2018
6. Rhee C, Strich JR, Klompas M, et al. SEP-1 has brought much needed attention to improving sepsis care…but now is the time to improve SEP-1. Crit Care Med. 2020; 48:779–782
7. Prescott HC, Iwashyna TJ. Improving sepsis treatment by embracing diagnostic uncertainty. Ann Am Thorac Soc. 2019; 16:426–429
8. Rhee C, Chiotos K, Cosgrove SE, et al. Infectious Diseases Society of America Position Paper: Recommended Revisions to the National Severe Sepsis and Septic Shock Early Management Bundle (SEP-1) Sepsis Quality Measure. Clin Infect Dis. 2020 May 6. [online ahead of print]
9. Kashiouris MG, Zemore Z, Kimball Z, et al. Supply chain delays in antimicrobial administration after the initial clinician order and mortality in patients with Sepsis. Crit Care Med. 2019; 47:1388–1395
10. Curtiss J, Karvetski C, Taylor B, et al. 1476: Intervals of delay in antibiotic administration for emergency department patients with sepsis. Crit Care Med. 2018; 46:722
11. 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; 315:762–774
    12. Rhee C, Zhang Z, Kadri SS, et al.; CDC Prevention Epicenters Program. Sepsis surveillance using adult sepsis events simplified eSOFA criteria versus sepsis-3 sequential organ failure assessment criteria. Crit Care Med. 2019; 47:307–314
    13. Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol. 2016; 183:758–764
    14. Dunnett CW. A multiple comparison procedure for comparing several treatments with a control. J Am Stat Assoc. 1955;50:1096–1121
    15. SSC Adult Guidelines. The Fourth Edition of "Surviving Sepsis Campaign: International Guidelines for Management of Sepsis and Septic Shock: 2016" Provides Guidance for the Clinician Caring for Adult Patients With Sepsis or Septic Shock. 2016Available at: Accessed October 21, 2020
    16. Taylor SP, Rozario N, Kowalkowski MA, et al. Trends in false-positive code sepsis activations in the emergency department. Ann Am Thorac Soc. 2020; 17:520–522
    17. Klein Klouwenberg PM, Cremer OL, van Vught LA, et al. Likelihood of infection in patients with presumed sepsis at the time of intensive care unit admission: A cohort study. Crit Care. 2015; 19:319
    18. Weinberger J, Rhee C, Klompas M. A critical analysis of the literature on time-to-antibiotics in suspected sepsis. J Infect Dis. 2020; 222:S110–S118

    antibiotics; emergency department; mortality; quality measures; sepsis

    Supplemental Digital Content

    Copyright © 2021 by the Society of Critical Care Medicine and Wolters Kluwer Health, Inc. All Rights Reserved.