Abstract
Rationale
Shorter time-to-antibiotics is lifesaving in sepsis, but programs to hasten antibiotic delivery may increase unnecessary antibiotic use and adverse events.
Objectives
We sought to estimate both the benefits and harms of shortening time-to-antibiotics for sepsis.
Methods
We conducted a simulation study using a cohort of 1,559,523 hospitalized patients admitted through the emergency department with meeting two or more systemic inflammatory response syndrome criteria (2013–2018). Reasons for hospitalization were classified as septic shock, sepsis, infection, antibiotics stopped early, and never treated (no antibiotics within 48 h). We simulated the impact of a 50% reduction in time-to-antibiotics for sepsis across 12 hospital scenarios defined by sepsis prevalence (low, medium, or high) and magnitude of “spillover” antibiotic prescribing to patients without infection (low, medium, high, or very high). Outcomes included mortality and adverse events potentially attributable to antibiotics (e.g., allergy, organ dysfunction, Clostridiodes difficile infection, and culture with multidrug-resistant organism).
Results
A total of 933,458 (59.9%) hospitalized patients received antimicrobial therapy within 48 hours of presentation, including 38,572 (2.5%) with septic shock, 276,082 (17.7%) with sepsis, 370,705 (23.8%) with infection, and 248,099 (15.9%) with antibiotics stopped early. A total of 199,937 (12.8%) hospitalized patients experienced an adverse event; most commonly, acute liver injury (5.6%), new MDRO (3.5%), and Clostridiodes difficile infection (1.7%). Across the scenarios, a 50% reduction in time-to-antibiotics for sepsis was associated with a median of 1 to 180 additional antibiotic-treated patients and zero to seven additional adverse events per death averted from sepsis.
Conclusions
The impacts of faster time-to-antibiotics for sepsis vary markedly across simulated hospital types. However, even in the worst-case scenario, new antibiotic-associated adverse events were rare.
Keywords: antibacterial agents, emergency service, hospital, cohort studies
Sepsis is a leading cause of hospitalization and contributes to one in every two to three hospital deaths in the United States (1, 2) Shorter time to antimicrobial therapy (i.e., time-to-antibiotics) is associated with reduced mortality in sepsis, particularly for patients who present with shock (3–5). For this reason, sepsis quality improvement initiatives have focused on reducing time-to-antibiotics. However, experts have cautioned that efforts to hasten sepsis treatment may result in increased antimicrobial use and associated harms, including among patients who turn out to have noninfectious causes for illness (6–8). However, there are limited empirical data on the burden of “spillover prescribing” (9)—additional patients being treated unnecessarily with antibiotics because of efforts to shorten the time-to-antibiotics for sepsis—or antibiotic-associated harms (10).
In a prior study of 152 Veterans Affairs (VA) and Kaiser Permanente Northern California (KPNC) hospitals, we showed that reductions in time-to-antibiotics for sepsis were not associated with increases in overall antimicrobial prescribing to patients who were plausibly at risk for sepsis (11). Rather, most hospitals simultaneously reduced time-to-antibiotics for sepsis and overall antimicrobial use in at-risk patients during the study period (11). However, it remains possible that efforts to hasten sepsis treatment could result in increased antimicrobial use and associated harms, particularly in hospitals with less robust stewardship programs.
In this study, we sought to estimate the incidence of antibiotic-associated harms in hospitalized patients and, using simulation, to quantify the potential positive and negative impacts of shortening time-to-antibiotics for sepsis in the setting of spillover prescribing. In the simulations, we considered 12 hypothetical scenarios defined by 1) hospital sepsis prevalence and 2) the magnitude of spillover prescribing.
Methods
Study Design and Cohort
This study included all adult patients hospitalized at 152 U.S. VA and KPNC hospitals from 2013 to 2018 who were admitted through the emergency department with two or more systemic inflammatory response syndrome (SIRS) criteria (12). Patient and hospitalization characteristics, including demographics, comorbidities, laboratory values, and time-to-antibiotics, were extracted from electronic health records as in prior work (11, 13–17). Time-to-antibiotics was measured as time from ED presentation to time of first administration of systemic antimicrobial therapy (15). Hospitalizations were classified into five mutually exclusive subtypes: septic shock, sepsis, infection, antimicrobial therapy stopped early, or no antimicrobial therapy administered within 48 hours (see Table E1 in the data supplement). Sepsis and septic shock hospitalizations were defined by objective evidence of acute organ dysfunction, initiation of systemic antimicrobial therapy within 48 hours, and administration of 4 or more consecutive days of antimicrobial therapy, similar to the Centers for Disease Control and Prevention’s definition of adult sepsis event (15, 16, 18).
Antibiotic-associated Adverse Events
We evaluated eight adverse events that have been described as potentially attributable to antimicrobial therapy: 1) allergic or hypersensitivity reaction; 2) acute thrombocytopenia (<150,000 cells per microliter and a >50% decline from baseline); 3) leukopenia (<4,500 cells per microliter and a >50% decline from baseline); 4) acute kidney injury (new renal replacement therapy or an increase of ⩾0.5 mg/dl and a >50% increase in serum creatinine from baseline on two or more measurements, consistent with the criteria for risk of renal dysfunction, injury to the kidney, failure of kidney function, loss of kidney function, and end-stage kidney disease [also known as RIFLE]) (19); 5) acute liver injury (alanine aminotransferase >80 IU/L and >50% increase from baseline or total bilirubin >2.4 mg/dl and a >50% increase from baseline); 6) Clostridiodes difficile infection (CDI; positive C. difficile testing and treatment with an anti–C. difficile antimicrobial); 7) new multidrug-resistant organism (MDRO) culture positivity (culture or swab newly positive for methicillin-resistant Staphylococcus aureus, vancomycin-resistant Enterococcus, carbapenem-resistant Enterobacteriaceae, extended-spectrum β-lactamase–producing Enterobacteriaceae, multidrug-resistant Pseudomonas species [a Pseudomonas that is resistant to at least one antibiotic in at least three different antibiotic classes], or Acinetobacter species) (11), and 8) new blood culture MDRO positivity. We additionally examined mortality at 30 and 90 days. All adverse events were identified through electronic health record data. Further information on the adverse event definitions is provided elsewhere (see Table E2). Conceptually, events that occurred within 48 hours of hospitalization were attributed to an underlying illness (except for allergic or hypersensitivity reaction), whereas events that occurred after 48 hours of hospitalization were considered plausibly related to treatment and are the focus of the study. Consistent with prior studies (10), the time window for the evaluation of adverse events varied across the outcomes, because the expected time frame for development varies across adverse events. For example, CDI and MDRO culture positivity were evaluated to 90 days, whereas allergic reaction was evaluated to 9 days after ED presentation.
Statistical Analysis
We fit a series of multivariable logistic regression models predicting each adverse event and adjusted for patient characteristics (age, sex, 30 Elixhauser chronic conditions, individual SIRS criteria, and individual acute organ dysfunctions on presentation). We report adjusted outcome probabilities in hospitalized patients who were treated versus not treated with antimicrobial therapy within 48 hours, and we present the adjusted odds ratios for the association of antimicrobial therapy with each outcome. We used these models to estimate the probability of 30-day mortality among sepsis hospitalizations with observed time-to-antibiotics versus shortened time-to-antibiotics and the probability of an adverse event with versus without receipt of antimicrobial therapy within 48 hours of ED arrival.
To estimate the impact of shortening time-to-antibiotics for sepsis (inclusive of resultant spillover prescribing and associated harms), we conducted a series of simulations in which time-to-antibiotics was reduced by 50% for hospitalized patients with sepsis or septic shock. We selected 50% as the maximal plausible effect to make clear what the tradeoffs may be. We completed simulations across 12 hospital types defined by the prevalence of sepsis (low, medium, or high prevalence) and the magnitude of spillover prescribing (low, medium, high, or very high spillover prescribing), because these vary by hospital and influence the impact of hastening time-to-antibiotics. Each simulated hospital contained 20,000 hospitalized patients who were admitted through the ED meeting two or more SIRS criteria, sampled at random from the full cohort without replacement. Sampling was stratified by hospitalization subtype (septic shock, sepsis, infection, etc.), with distributions for low, medium, and high sepsis prevalence informed by the pooled proportions of septic shock, sepsis, and so forth, in the bottom 26, middle 100, and top 26 hospitals in our cohort, as ranked by proportion of hospitalized patients with septic shock. Although prior work in this cohort exhibited no evidence for spillover prescribing when shortening time-to-antibiotics for sepsis (11), we simulated a wide range in the rate of spillover prescribing to understand the impacts of shortening time-to-antibiotics under hypothetical conditions with ineffective antimicrobial stewardship. Low spillover prescribing was set at 0.2% per hour reduction in median time-to-antibiotics, as measured empirically in the cohort (11) (i.e., 0.2% of untreated hospitalized patients were newly treated with antibiotics per hour reduction in median time-to-antibiotics). Cases of medium, high, and very high spillover prescribing were set at 1%, 5%, and 10%, respectively, of hospitalized patients who were newly treated with antibiotics per hour reduction in median time-to-antibiotics for sepsis (5, 25, and 50 times the observed spillover) to yield worst-case estimates.
For each scenario, we completed 1,000 simulations to quantify the positive impact of faster time-to-antibiotics for sepsis (number of deaths from sepsis averted) and the negative impacts (number of patients for whom antibiotics had been additionally administered within 48 hours, number of patients with new adverse events resulting from antibiotics) with different hospital populations. For each simulation, we performed the following steps: 1) We sample 20,000 patients from the overall cohort into the hypothetical hospital; 2) we estimated the impact of a 50% reduction in time-to-antibiotics on 30-day mortality among hospitalized patients with sepsis and septic shock by comparing mean predicted probabilities for mortality with the observed time-to-antibiotics versus a 50% shorter time-to-antibiotics; 3) we calculated the number of hospitalized patients who would be newly treated with antibiotics on the basis of the spillover rate, the number of untreated hospitalized patients, and the absolute reduction in median time-to-antibiotics; and 4) we estimated the number of new adverse events among the hospitalized patients who were newly treated with antibiotics by comparing mean predicted probability for each adverse event if treated versus not treated with antibiotics. We report the median number of sepsis deaths averted, the median number of hospitalized patients who were newly treated with antibiotics, and the median number of new adverse events stemming from new antibiotic treatment across simulations. We also report the median number of hospitalized patients who were newly treated with antibiotics per sepsis death averted for each scenario. Furthermore, we report the distribution of these values across the 1,000 simulations for each scenario (see the data supplement). To facilitate understanding and replication of our simulations, we refer readers to the statistical code, which is publicly available at GitHub (https://github.com/CCMRcodes/AntibioticTimingSimulations).
Data management was performed using SAS 9.4 (SAS Institute), and analyses were performed using Stata 16 (StataCorp) and R (R Core Team and R Foundation for Statistical Computing). This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (or, STROBE) reporting guideline (20).
Institutional Review Board Approval
The study was reviewed by the University of Michigan, VA Ann Arbor, and KPNC Institutional Review Boards and was deemed exempt from the need for consent under 45 CFR §46, category 4 (secondary use of identifiable data).
Results
Population Characteristics
From 2013 to 2018, there were 1,559,523 hospitalizations for patients admitted through the emergency department with meeting two or more SIRS criteria at 152 VA or KPNC hospitals, thereby meeting inclusion criteria for the study (see Table E3). Cohort characteristics are presented in Table 1. The median age was 67 years (interquartile range, 59–77), 30.5% had comorbid chronic kidney disease, 17.7% had comorbid liver disease, and 32.4% had comorbid heart failure. Systemic antimicrobial therapy was administered to 933,458 (59.9%) hospitalized patients within 48 hours of emergency department presentation, including 38,572 (2.5%) with septic shock, 276,082 (17.7%) with sepsis, 370,705 (23.8%) with infection, and 248,099 (15.9%) with antimicrobial therapy stopped early. The most common sites of infection among patients with sepsis were pulmonary (35.5%) and genitourinary (17.9%) (see Table E4). The proportion of hospitalized patients with septic shock differed meaningfully across the bottom 26, middle 100, and top 26 hospitals (0.3%, 1.8%, and 4.2%, respectively), as did the proportion with sepsis (11.6%, 16.3%, and 21.3%, respectively) (see Tables E5 and E6).
Table 1.
Characteristics of hospitalized patients overall and by subgroup
| Characteristic | Overall (n = 1,559,523) | Treatment with Antimicrobial Therapy within 48 h of Arrival to ED |
Untreated (n = 626,065) | |||
|---|---|---|---|---|---|---|
| Septic Shock (n = 38,572) | Sepsis (n = 276,082) | Infection (n = 370,705) | Stopped Early (n = 248,099) | |||
| Age, median (IQR) | 67 (59–77) | 69 (61–78) | 69 (62–79) | 68 (59–78) | 68 (59–79) | 66 (58–75) |
| SIRS criteria on presentation, n (%) | ||||||
| Abnormal WBC | 884,428 (56.7) | 32,019 (83.0) | 203,938 (73.9) | 238,652 (64.4) | 148,165 (59.7) | 261,654 (41.8%) |
| Abnormal temperature | 747,934 (48.0) | 24,389 (63.2) | 153,957 (55.8) | 185,317 (50.0) | 108,655 (43.8) | 275,616 (44.0%) |
| Elevated heart rate | 1,358,819 (87.1) | 35,402 (91.8) | 245,668 (89.0) | 326,019 (88.0) | 215,983 (87.1) | 535,747 (85.6%) |
| Elevated respiratory rate | 1,042,064 (66.8) | 33,657 (87.3) | 193,489 (70.1) | 241,864 (65.2) | 168,457 (67.9) | 404,597 (64.6%) |
| Acute organ dysfunction on presentation, n (%) | ||||||
| Renal | 335,804 (21.5) | 22,160 (57.5) | 143,779 (52.1) | 0 (0) | 53,499 (21.6) | 116,366 (18.6) |
| Lactate elevation | 268,897 (17.2) | 26,232 (68.0) | 144,305 (52.3) | 0 (0) | 47,018 (19.0) | 51,342 (8.2) |
| Hematological | 80,408 (5.2) | 7,197 (18.7) | 34,971 (12.7) | 0 (0) | 11,751 (4.7) | 26,489 (4.2) |
| Hepatic | 68,779 (4.4) | 4,610 (12.0) | 30,916 (11.2) | 0 (0) | 10,291 (4.2) | 22,962 (3.7) |
| Mechanical ventilation | 41,820 (2.7) | 15,498 (40.2) | 13,707 (5.0) | 0 (0) | 5,529 (2.2) | 7,086 (1.1) |
| Shock | 52,971 (3.4) | 38,572 (100) | 0 (0) | 0 (0) | 6,392 (2.6) | 8,007 (1.3) |
| Chronic conditions, n (%)* | ||||||
| Renal disease | 475,230 (30.5) | 14,910 (38.7) | 105,667 (38.3) | 88,847 (24.0) | 80,285 (32.4) | 185,521 (29.6) |
| Liver disease | 275,630 (17.7) | 10,232 (26.5) | 57,692 (20.9) | 48,295 (13.0) | 43,413 (17.5) | 115,998 (18.5) |
| Heart failure | 505,061 (32.4) | 15,602 (40.5) | 90,560 (32.8) | 100,658 (27.2) | 78,229 (31.5) | 220,012 (35.1) |
| Pulmonary disease | 700,590 (44.9) | 17,135 (44.4) | 126,699 (45.9) | 193,081 (52.1) | 115,043 (46.4) | 248,632 (39.7) |
Definition of abbreviations: ED = emergency department; IQR = interquartile range; SIRS = systemic inflammatory response syndrome; WBC = white blood cell.
Comorbidities were defined on the basis of diagnosis codes using the Elixhauser definitions, with a 1.5-yr lookback period from the date of emergency department admission. Acute organ dysfunction was identified on the basis of electronic health record data, as in prior work (17).
Incidence of Adverse Events
Adverse events are presented in Table 2. Overall, 12.8% of hospitalized patients experienced an adverse event of interest; most commonly, acute liver injury (5.6%), new MDRO culture positivity (3.5%), and CDI (1.7%). Adverse events occurred in 28.5% of hospitalized patients with septic shock, 19.5% of those with sepsis, 12.2% of those with infection, 11.3% of those with antimicrobial therapy stopped early, and 9.9% of untreated hospitalized patients. In multivariable models adjusted for patient characteristics, antimicrobial therapy was associated with increased risk for allergy or hypersensitivity reaction (adjusted odds ratio [aOR], 4.3; 95% confidence interval [CI], 4.0–4.5), leukopenia (aOR, 1.5; 95% CI, 1.4–1.5), acute kidney injury (aOR, 1.4; 95% CI, 1.3–1.4), CDI (aOR, 2.1; 95% CI, 2.1–2.2), new MDRO culture positivity (aOR, 2.6; 95% CI, 2.5–2.6), and new MDRO blood culture positivity (aOR, 2.1; 95% CI, 2.0–2.2), but not with thrombocytopenia (aOR, 1.0; 95% CI, 0.9–1.0) or acute liver injury (aOR, 0.9; 95% CI, 0.9–0.9).
Table 2.
Study outcomes overall and by hospitalization type
| Study Outcome | Overall Cohort (n = 1,559,523) |
Treatment with Antimicrobial Therapy within 48 h |
Untreated (n = 626,065) | Adjusted % |
aOR (95% CI) for Antimicrobial Therapy | ||||
|---|---|---|---|---|---|---|---|---|---|
| Septic Shock (n = 38,572) |
Sepsis (n = 276,082) |
Infection (n = 370,705) |
Stopped Early (n = 248,099) |
Untreated | Treated | ||||
| Adverse events, n (%) | |||||||||
| Allergy | 10,136 (0.7) | 421 (1.1) | 2,879 (1.0) | 3,711 (1.0) | 1,792 (0.7) | 1,333 (0.2) | 0.2 | 0.9 | 4.3 (4.0–4.5) |
| Thrombocytopenia | 15,286 (1.0) | 2,413 (6.3) | 4,689 (1.7) | 1,749 (0.5) | 1,839 (0.7) | 4,596 (0.7) | 1.0 | 1.0 | 1.0 (0.9–1.0) |
| Leukopenia | 14,100 (0.9) | 933 (2.4) | 4,648 (1.7) | 2,850 (0.8) | 1,826 (0.7) | 3,843 (0.6) | 0.7 | 1.0 | 1.5 (1.4–1.5) |
| Acute kidney injury | 26,415 (1.7) | 2,386 (6.2) | 7,397 (2.7) | 5,886 (1.6) | 2,965 (1.2) | 7,781 (1.2) | 1.4 | 1.9 | 1.4 (1.3–1.4) |
| Acute liver injury | 88,010 (5.6) | 4,373 (11.3) | 19,966 (7.2) | 14,932 (4.0) | 12,097 (4.9) | 36,642 (5.9) | 5.9 | 5.5 | 0.9 (0.9–0.9) |
| C. difficile infection | 26,030 (1.7) | 1,540 (4.0) | 8,640 (3.1) | 6,643 (1.8) | 3,546 (1.4) | 5,661 (0.9) | 1.0 | 2.1 | 2.1 (2.1–2.2) |
| MDRO culture positivity | 55,143 (3.5) | 2,963 (7.7) | 17,114 (6.2) | 15,547 (4.2) | 8,249 (3.3) | 11,270 (1.8) | 1.9 | 4.6 | 2.6 (2.5–2.6) |
| MDRO blood culture | 7,540 (0.5) | 521 (1.4) | 2,748 (1.0) | 1,431 (0.4) | 1,155 (0.5) | 1,685 (0.3) | 0.3 | 0.6 | 2.1 (2.0–2.2) |
| Any adverse event | 199,937 (12.8) | 11,009 (28.5) | 53,761 (19.5) | 45,173 (12.2) | 27,961 (11.3) | 62,033 (9.9) | 10.3 | 14.4 | 1.5 (1.5–1.5) |
| Mortality, n (%) | |||||||||
| 30-d mortality | 121,690 (7.8) | 13,126 (34.0) | 31,733 (11.5) | 18,976 (5.1) | 21,001 (8.5) | 36,854 (5.9) | 7.8 | 7.8 | 1.0 (1.0–1.0) |
| 90-d mortality | 216,101 (13.9) | 15,914 (41.3) | 52,983 (19.2) | 38,334 (10.3) | 37,279 (15.0) | 71,591 (11.4) | 13.7 | 13.9 | 1.0 (1.0–1.0) |
Definition of abbreviations: aOR = adjusted odds ratio; CI = confidence interval; MDRO = multidrug-resistant organism.
Multivariable logistic regression model adjusted for patient age, sex, 30 chronic conditions, individual SIRS criteria, and individual acute organ dysfunctions. Adverse events defined based on data from the electronic health record based on methods described in Table E2.
Among the overall cohort, 30-day mortality was 7.8%, and 90-day mortality was 13.9%. Mortality differed substantially by subtype, with 30-day mortality ranging from 5.1% for hospitalized patients with SIRS-positive infection to 34.0% for hospitalized patients with septic shock and 90-day mortality ranging from 10.3% for hospitalized patients with infection to 41.3% for hospitalized patients with septic shock (Table 2).
Simulation Results
Median simulation results are presented in Tables 3 and 4. The distributions of results within each scenario are presented elsewhere (see Table E7) and were generally narrow relative to the differences in median estimates across scenarios. Each simulated hospital included 20,000 SIRS-positive hospitalized patients drawn at random from the overall cohort. Simulated hospitals with low, medium, and high sepsis prevalence had 2,377 (11.9%), 3,624 (18.1%), and 5,099 (25.5%) hospitalized patients with sepsis or septic shock, respectively (Table 3). Median time-to-antibiotics for sepsis in hospitals with low, medium, and high sepsis prevalence were 222, 218, and 215 minutes, respectively. Under the simulated 50% reduction, time-to-antibiotics decreased by 111 minutes, 109 minutes, and 107 minutes, respectively (see Figure E1). Reducing time-to-antibiotics by 50% resulted in a median of 9, 14, and 20 sepsis deaths averted among hospitals with low, medium, and high sepsis prevalence, respectively, or absolute risk reductions for 30-day mortality in hospitalized patients with sepsis of 0.38%, 0.39%, and 0.41%, respectively.
Table 3.
Simulation results for hospitalized patients with sepsis and septic shock
| Variable | Sepsis Prevalence in Hospitals |
||
|---|---|---|---|
| Low | Medium | High | |
| Hospitalized patients with sepsis or septic shock, n | 2,377 | 3,624 | 5,099 |
| Observed outcomes among hospitalized patients with sepsis or septic shock | |||
| Median time-to-antibiotics for sepsis, min for median simulation | 222 | 218 | 215 |
| 30-d mortality, n for median simulation | 288 | 497 | 774 |
| 30-d mortality, % for median simulation | 12.1 | 13.7 | 15.2 |
| Outcomes among hospitalized patients with sepsis or septic shock with 50% reduction in time-to-antibiotics | |||
| Median time-to-antibiotics for sepsis, min for median simulation | 111 | 109 | 108 |
| 30-d mortality, n for median simulation | 279 | 483 | 753 |
| 30-d mortality, % for median simulation | 11.7 | 13.3 | 14.8 |
| Sepsis deaths averted in the median simulation, n | 9 | 14 | 20 |
| Absolute sepsis mortality risk reduction, % | 0.38 | 0.39 | 0.41 |
Each simulated hospital contained N = 20,000 patients who were admitted through the emergency department with positive systemic inflammatory response syndrome criteria, suggestive of potential sepsis. The proportions of systemic inflammatory response syndrome–positive emergency department admissions with septic shock, sepsis, infection, antibiotic treatment stopped early (i.e., infection was ruled out), and no antibiotic treatment within 48 hours were defined for the low, medium, and high sepsis prevalence simulated hospitals on the basis of the distributions observed in the bottom 26, middle 100, and top 26 hospitals, respectively, in our cohort, as ranked by sepsis prevalence. The low, medium, and high sepsis prevalence simulated hospitals had 66, 358, and 832 septic shock hospitalizations, respectively; and 2,311, 3,266, and 4,267 severe sepsis hospitalizations, respectively. The numbers presented here are the median values for n = 1,000 simulations completed for each scenario.
Table 4.
Median simulation results, including spillover antibiotic prescribing and associated adverse events
| Sepsis Prevalence/Spillover Prescribing Scenario* | No. of Sepsis Deaths Averted | No. of Hospitalized Patients Newly Treated with Antibiotics | No. of Additional Adverse Events Caused by Spillover Prescribing of Antibiotics |
|||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Any Adverse Event | Allergic Reaction | Thrombo cytopenia | Leuko penia | Acute Kidney Injury | Acute Liver Injury | C. difficile Infection | MDRO on Any Culture | MDRO on Blood Culture | 90-d Mortality | |||
| Low/low | 9 | 32 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 |
| Medium/low | 14 | 32 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 |
| High/low | 20 | 22 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 0 |
| Low/medium | 9 | 162 | 6 | 1 | 0 | 0 | 1 | 0 | 2 | 4 | 0 | 0 |
| Medium/medium | 14 | 162 | 7 | 1 | 0 | 0 | 1 | 0 | 2 | 4 | 0 | 0 |
| High/medium | 20 | 109 | 4 | 1 | 0 | 0 | 0 | 0 | 1 | 3 | 0 | 0 |
| Low/high | 9 | 810 | 33 | 6 | 0 | 2 | 4 | 0 | 8 | 21 | 2 | 2 |
| Medium/high | 14 | 810 | 33 | 6 | 0 | 2 | 4 | 0 | 8 | 22 | 2 | 2 |
| High/high | 20 | 547 | 22 | 4 | 0 | 2 | 2 | 0 | 5 | 15 | 2 | 1 |
| Low/very high | 9 | 1,620 | 65 | 11 | 0 | 4 | 7 | 0 | 16 | 43 | 5 | 3 |
| Medium/very high | 14 | 1,620 | 65 | 11 | 0 | 4 | 7 | 0 | 16 | 43 | 5 | 3 |
| High/very high | 20 | 1,094 | 44 | 8 | 0 | 3 | 5 | 0 | 11 | 29 | 3 | 2 |
Definition of abbreviation: MDRO = multidrug-resistant organism.
The simulation scenarios were defined by sepsis prevalence and magnitude of spillover prescribing. In hospitals with low, medium, and high prevalence, the proportions of SIRS-positive hospitalized patients with sepsis or septic shock were set at 11.9%, 18.1%, and 25.5%, respectively, on the basis of observed distributions in our cohort. In hospitals with low, medium, high, and very high spillover prescribing, the magnitudes of additional antibiotic-treated patients were set at 0.2%, 1%, 5%, and 10% of untreated patients being newly treated with antibiotics per hour reduction in median time-to-antibiotics. In our prior work, the point estimate for spillover was 0.2%, which we set as low spillover prescribing. Cases of medium, high, and very high spillover prescribing were set at 5, 25, and 50 times the observed spillover to simulate worse-case scenarios.
The median number of additional hospitalized patients who were treated with antimicrobial therapy varied widely across scenarios, from 22 additional hospitalized patients (high sepsis prevalence, low spillover prescribing) to 1,620 additional hospitalized patients (low sepsis prevalence, very high spillover prescribing) (Table 4). The median number of hospitalized patients with new adverse events related to new antimicrobial therapy ranged from 1 per 20,000 simulated hospitalizations (high sepsis prevalence, low spillover prescribing) to 65 per 20,000 simulated hospitalizations (low sepsis prevalence, very high spillover prescribing) in each scenario. The most common adverse events were new MDRO culture positivity (occurring in a median of 1–43 hospitalized patients across scenarios) and CDI (occurring in a median of 0–16 hospitalized patients across simulations). Across all scenarios, new thrombocytopenia, leukopenia, acute liver injury, and mortality events were rare (four hospitalized patients or fewer). The median number of additional hospitalized patients who were treated with antimicrobial therapy per sepsis death averted ranged from 1 (high sepsis prevalence, low spillover prescribing) to 180 (low sepsis prevalence, very high spillover prescribing) across scenarios, whereas the number of patients with new, antibiotic-associated adverse event per death averted ranged from 0 to 7 (Table 5).
Table 5.
Data on patients newly treated with antibiotics and adverse events per sepsis death averted across simulations
| Sepsis Prevalence/Spillover Prescribing Scenarios* | No. of Sepsis Deaths Averted | No. of Hospitalized Patients Newly Treated with Antibiotics | No. of Hospitalized Patients Newly Treated with Antibiotics per Sepsis Death Averted | No. of Hospitalized Patients with New Antibiotic-associated Adverse Event per Sepsis Death Averted |
|---|---|---|---|---|
| Low/low | 9 | 32 | 4 | 0 |
| Medium/low | 14 | 32 | 2 | 0 |
| High/low | 20 | 22 | 1 | 0 |
| Low/medium | 9 | 162 | 18 | 1 |
| Medium/medium | 14 | 162 | 12 | 1 |
| High/medium | 20 | 109 | 5 | 0 |
| Low/high | 9 | 810 | 90 | 4 |
| Medium/high | 14 | 810 | 58 | 2 |
| High/high | 20 | 547 | 27 | 1 |
| Low/very high | 9 | 1,620 | 180 | 7 |
| Medium/very high | 14 | 1,620 | 116 | 5 |
| High/very high | 20 | 1,094 | 55 | 2 |
The simulation scenarios were defined by sepsis prevalence and magnitude of spillover prescribing. In low-, medium-, and high-prevalence hospitals, the proportions of SIRS-positive hospitalized patients with sepsis or septic shock were set at 11.9%, 18.1%, and 25.5%, respectively, on the basis of observed distributions in our cohort. In hospitals with low, medium, high, and very high spillover prescribing, the magnitude of additional antibiotic-treated patients was set at 0.2%, 1%, 5%, and 10%, respectively, of untreated patients being newly treated with antibiotics per hour reduction in median time-to-antibiotics. In our prior work, the point estimate for spillover was 0.2%, which we set as low spillover prescribing. Cases of medium, high, and very high spillover prescribing were set at 5, 25, and 50 times the observed spillover to simulate worse-case scenarios.
Discussion
In this study of over 1.5 million patients with potential infection hospitalized across 152 hospitals, we report for the first time on a comprehensive panel of adverse events, we measure the impact of antimicrobial therapy on these outcomes, and we use these empirical data to estimate population-level trade-offs that might occur when time-to-antibiotic reductions exacerbate spillover prescribing. We show that adverse events potentially attributable to antimicrobial therapy occur in approximately one in eight hospitalizations. We demonstrate that the impact of shortening time-to-antibiotics on antimicrobial use and adverse events varies markedly, depending on the prevalence of sepsis and the magnitude of spillover prescribing within a hospital. In the setting of low spillover prescribing, consistent with the level measured empirically in real-world data (11), a 50% reduction in time-to antibiotics for sepsis was associated with only one to four patients newly treated with antibiotics per sepsis death averted and no additional adverse events. In the worst-case scenario of very high spillover prescribing (50-fold higher than observed in our prior study), a 50% reduction in time-to-antibiotics for sepsis was associated with 55 to 180 patients newly treated with antibiotics and two to seven new adverse events per sepsis death averted, depending on the prevalence of sepsis in the hospital.
Our prior study showed that shortening time-to-antibiotics was not associated with increases in antimicrobial prescribing in this cohort (11). However, it remains possible that efforts to shorten time-to-antibiotics for sepsis could result in additional antimicrobial prescribing and associated adverse events. In this study, we show that—even under simulations with very high spillover prescribing—the benefits of shortening time-to-antibiotics appear to handily outweigh the harms. However, the number of additional antibiotic-treated patients varies markedly according to the magnitude of spillover prescribing, demonstrating the importance of promoting and monitoring antimicrobial stewardship alongside efforts to hasten antimicrobial delivery in sepsis. These results fill a critical knowledge gap regarding the potential tradeoffs of shortening time-to-antibiotics in sepsis (6–8). Furthermore, the results argue against the suggestion that antimicrobial timing targets should be abandoned in sepsis because of their potential for adverse effects. However, they support the need for antimicrobial use monitoring alongside efforts to reduce time-to-antibiotics in sepsis (11, 21).
Our findings on the prevalence of adverse events expand on the limited existing prior literature. In a study of 1,488 patients hospitalized in 2013–2014 and treated with systemic antimicrobial therapy, 1 in 5 patients experienced at least one antibiotic-associated adverse event (most commonly, gastrointestinal or renal adverse events) on the basis of adjudication by two clinicians trained in infectious diseases (10). Our findings were in a similar range, with 14.8% of antibiotic-treated, SIRS-positive patients experiencing possible treatment-related adverse events. Our study sample included a far larger, multihospital cohort and compared adverse event rates in antimicrobial-treated versus untreated patients using an intention-to-treat framework in which antimicrobial treatment was defined by treatment within 48 hours of presentation to estimate the association of antimicrobial therapy on adverse events. We show that antimicrobial therapy was most strongly associated with the development of allergic reactions, MDRO culture positivity, and CDI and that the absolute increase in adverse events was greatest for MDRO culture positivity, CDI, and allergic reactions.
Our findings on the mortality benefit of shorter time-to-antibiotics are also similar to findings from prior studies. Our simulations showed that reduction in time-to-antibiotics by around 100 minutes was associated with a 0.4% absolute reduction in 30-day mortality among patients with sepsis and septic shock. These results are similar to those of the The Prehospital ANTibiotics Against Sepsis (or, PHANTASi) trial, in which patients who were randomized to prehospital antibiotics received antibiotics 96 minutes faster than the usual-care arm and had a 0.4% absolute reduction in 28-day mortality (albeit nonsignificant) (22). It is interesting that simulating an average decrease in time-to-antibiotics of relatively large magnitude was associated with only modest increases in sepsis deaths averted, which suggests that additional approaches to reducing mortality beyond antibiotic timing are needed to combat the impact of sepsis.
Our findings should be interpreted in the context of some limitations. First, given the size of the study population, clinician adjudication of adverse events was not feasible. It is, therefore, likely that some adverse events may have been due to underlying illness and not the direct result of antimicrobial therapy. If anything, our approach is likely to have overestimated the number of new adverse events attributable to antibiotics in our simulations, which relied on estimates of association between antimicrobial therapy and adverse events. Second, there are limited data to inform the highest plausible level of spillover prescribing that may result from shortening time-to-antibiotics in sepsis. We tested very high magnitudes of spillover prescribing—up to 50 times the levels observed in our prior study (11)—to generate worst-case-scenario estimates of harm. Third, our simulations used data from hospitals with strong stewardship programs to simulate what might happen in scenarios of high spillover prescribing. The real-world data were used to measure the association of antibiotic treatment with adverse events, whereas the rate of antibiotic prescribing was set in the simulations. The simulations assume that the spillover prescribing would be of similar duration and broadness of existing antibiotic treatment. Our simulations did not consider any differences in duration or antibacterial spectrum that could accompany spillover prescribing or that may either increase or decrease the magnitude of harms estimated in our simulations. However, in prior work, we found no evidence that shortening time-to-antibiotics for sepsis was associated with increases in either duration or antibacterial spectrum (11). Fourth, although we examined eight important antibiotic-associated harms, these outcomes were not exhaustive. We were not able to measure all possible adverse events or potential long-term consequences related to microbiome changes. Fifth, although we used recommended electronic health record–based criteria to define sepsis and septic shock, there is a possibility for misclassification in both directions, including false-positive identification of sepsis in patients receiving chronic antibiotic therapy and hospitalized for other causes. Sixth, this study did not evaluate the impact of specific antimicrobial stewardship practices on spillover prescribing, which is a topic warranting further study. Seventh, to elucidate potential benefits and harms, we chose a significant relative change in antimicrobial timing for simulations. However, a 50% reduction may be difficult to achieve in practice; both benefits and harms would be more modest with smaller reductions in antimicrobial timing. Finally, although our simulations used statistical models to quantify potential impact on adverse events, they may not have accurately replicated the true relationship between antibiotic timing and outcomes.
Conclusions
In this study of more than 1.5 million patients hospitalized with potential infection at 152 U.S. hospitals, approximately 1 in 8 experienced an adverse event plausibly attributable to antimicrobial therapy. In simulations testing the impact of a 50% reduction in time-to-antibiotics for sepsis, the number of patients newly treated with antibiotics varied widely depending on the prevalence of sepsis and the magnitude of spillover prescribing, demonstrating the importance of monitoring antimicrobial stewardship concurrent to encouraging faster sepsis treatment. Even under the worst case scenario, however, the benefits of shortening time-to-antibiotics for sepsis outweighed the harms.
Acknowledgments
Acknowledgment
The authors thank Xiao Qing Wang, formerly of the University of Michigan; and Mahesh Bubule, Jonathan Lontok, and Mei-Tsung Lee of the KPNC Division of Research, who assisted with database creation and management. The authors also thank Cainnear Hogan of the VA Center for Clinical Management Research and Fernando Berreda of the KPNC Division of Research, who provided administrative and regulatory support.
Footnotes
Supported by the Agency for Healthcare Research and Quality grant R01 HS026725 (to H.C.P. and V.X.L.); the U.S. Department of Veterans Affairs, Health Services Research and Development Service grant IIR 20-313 (to H.C.P.); and the National Institute of General Medical Sciences grant R35GM128672 (to V.X.L.). J.P.D. was supported by National Heart, Lung, and Blood Institute grant K12 HL138039. This work represents the views of the authors and does not necessarily represent the views of the Department of Veterans Affairs or the U.S. government.
Author Contributions: All authors made substantial contributions to the conception or design of the work or the acquisition, analysis, or interpretation of data. J.P.D. and H.C.P. drafted the manuscript. All authors revised it critically for intellectual content. All authors approved the manuscript for submission and agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.
This article has a data supplement, which is accessible from this issue’s table of contents at www.atsjournals.org.
Author disclosures are available with the text of this article at www.atsjournals.org.
References
- 1. Liu V, Escobar GJ, Greene JD, Soule J, Whippy A, Angus DC, et al. Hospital deaths in patients with sepsis from 2 independent cohorts. JAMA . 2014;312:90–92. doi: 10.1001/jama.2014.5804. [DOI] [PubMed] [Google Scholar]
- 2. Rhee C, Jones TM, Hamad Y, Pande A, Varon J, O’Brien C, et al. Centers for Disease Control and Prevention (CDC) Prevention Epicenters Program Prevalence, underlying causes, and preventability of sepsis-associated mortality in US acute care hospitals. JAMA Netw Open . 2019;2:e187571. doi: 10.1001/jamanetworkopen.2018.7571. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Liu VX, Fielding-Singh V, Greene JD, Baker JM, Iwashyna TJ, Bhattacharya J, et al. The timing of early antibiotics and hospital mortality in sepsis. Am J Respir Crit Care Med . 2017;196:856–863. doi: 10.1164/rccm.201609-1848OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Peltan ID, Brown SM, Bledsoe JR, Sorensen J, Samore MH, Allen TL, et al. ED door-to-antibiotic time and long-term mortality in sepsis. Chest . 2019;155:938–946. doi: 10.1016/j.chest.2019.02.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Seymour CW, Gesten F, Prescott HC, Friedrich ME, Iwashyna TJ, Phillips GS, et al. Time to treatment and mortality during mandated emergency care for sepsis. N Engl J Med . 2017;376:2235–2244. doi: 10.1056/NEJMoa1703058. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Klompas M, Calandra T, Singer M. Antibiotics for sepsis—finding the equilibrium. JAMA . 2018;320:1433–1434. doi: 10.1001/jama.2018.12179. [DOI] [PubMed] [Google Scholar]
- 7. Rhee C, Heil EL. Toward more rational time-to-antibiotic metrics for suspected sepsis. Crit Care Med . 2021;49:858–860. doi: 10.1097/CCM.0000000000004905. [DOI] [PubMed] [Google Scholar]
- 8. Singer M. Antibiotics for sepsis: does each hour really count, or is it incestuous amplification? Am J Respir Crit Care Med . 2017;196:800–802. doi: 10.1164/rccm.201703-0621ED. [DOI] [PubMed] [Google Scholar]
- 9. Francetic I, Meacock R, Elliott J, Kristensen SR, Britteon P, Lugo-Palacios DG, et al. Framework for identification and measurement of spillover effects in policy implementation: intended non-intended targeted non-targeted spillovers (INTENTS) Implement Sci Commun . 2022;3:30. doi: 10.1186/s43058-022-00280-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Tamma PD, Avdic E, Li DX, Dzintars K, Cosgrove SE. Association of adverse events with antibiotic use in hospitalized patients. JAMA Intern Med . 2017;177:1308–1315. doi: 10.1001/jamainternmed.2017.1938. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Prescott HC, Seelye S, Wang XQ, Hogan CK, Smith JT, Kipnis P, et al. Temporal trends in antimicrobial prescribing during hospitalization for potential infection and sepsis. JAMA Intern Med . 2022;182:805–813. doi: 10.1001/jamainternmed.2022.2291. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. American College of Chest Physicians/Society of Critical Care Medicine Consensus Conference: definitions for sepsis and organ failure and guidelines for the use of innovative therapies in sepsis. Crit Care Med . 1992;20:864–874. [PubMed] [Google Scholar]
- 13. Vincent BM, Wiitala WL, Burns JA, Iwashyna TJ, Prescott HC. Using Veterans Affairs Corporate Data Warehouse to identify 30-day hospital readmissions. Health Serv Outcomes Res Methodol . 2018;18:143–154. doi: 10.1007/s10742-018-0178-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Wang XQ, Vincent BM, Wiitala WL, Luginbill KA, Viglianti EM, Prescott HC, et al. Veterans Affairs patient database (VAPD 2014-2017): building nationwide granular data for clinical discovery. BMC Med Res Methodol . 2019;19:94. doi: 10.1186/s12874-019-0740-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Wayne MT, Seelye S, Molling D, Wang XQ, Donnelly JP, Hogan CK, et al. Temporal trends and hospital variation in time-to-antibiotics among veterans hospitalized with sepsis. JAMA Netw Open . 2021;4:e2123950. doi: 10.1001/jamanetworkopen.2021.23950. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Wayne MT, Molling D, Wang XQ, Hogan CK, Seelye S, Liu VX, et al. Measurement of sepsis in a national cohort using three different methods to define baseline organ function. Ann Am Thorac Soc . 2021;18:648–655. doi: 10.1513/AnnalsATS.202009-1130OC. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Karlic KJ, Clouse TL, Hogan CK, Garland A, Seelye S, Sussman JB, et al. Comparison of administrative versus electronic health record-based methods for identifying sepsis hospitalizations. Ann Am Thorac Soc . 2023;20:1309–1315. doi: 10.1513/AnnalsATS.202302-105OC. [DOI] [PubMed] [Google Scholar]
- 18.Hospital toolkit for adult sepsis surveillance. Atlanta, GA: Centers for Disease Control and Prevention; 2018. https://www.cdc.gov/sepsis/pdfs/sepsis-surveillance-toolkit-mar-2018_508.pdf [Google Scholar]
- 19. Bellomo R, Ronco C, Kellum JA, Mehta RL, Palevsky P, Acute Dialysis Quality Initiative workgroup Acute renal failure—definition, outcome measures, animal models, fluid therapy and information technology needs: the Second International Consensus Conference of the Acute Dialysis Quality Initiative (ADQI) Group. Crit Care . 2004;8:R204–R212. doi: 10.1186/cc2872. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP, STROBE Initiative The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Ann Intern Med . 2007;147:573–577. doi: 10.7326/0003-4819-147-8-200710160-00010. [DOI] [PubMed] [Google Scholar]
- 21. Prescott HC, Iwashyna TJ. Improving sepsis treatment by embracing diagnostic uncertainty. Ann Am Thorac Soc . 2019;16:426–429. doi: 10.1513/AnnalsATS.201809-646PS. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Alam N, Oskam E, Stassen PM, Exter PV, van de Ven PM, Haak HR, et al. PHANTASi Trial Investigators and the ORCA (Onderzoeks Consortium Acute Geneeskunde) Research Consortium the Netherlands Prehospital antibiotics in the ambulance for sepsis: a multicentre, open label, randomised trial. Lancet Respir Med . 2018;6:40–50. doi: 10.1016/S2213-2600(17)30469-1. [DOI] [PubMed] [Google Scholar]