Research

JAMA Internal Medicine | Original Investigation | LESS IS MORE

Mortality of Patients With Sepsis

Administered Piperacillin-Tazobactam vs Cefepime
Rishi Chanderraj, MD, MSc; Andrew J. Admon, MD, MSc, MPH; Ying He, PhD; Mark Nuppnau, MSc;
Owen R. Albin, MD; Hallie C. Prescott, MD, MSc; Robert P. Dickson, MD; Michael W. Sjoding, MD, MSc

Supplemental content
IMPORTANCE Experimental and observational studies have suggested that empirical
treatment for bacterial sepsis with antianaerobic antibiotics (eg, piperacillin-tazobactam)
is associated with adverse outcomes compared with anaerobe-sparing antibiotics (eg,
cefepime). However, a recent pragmatic clinical trial of piperacillin-tazobactam and cefepime
showed no difference in short-term outcomes at 14 days. Further studies are needed to help
clarify the empirical use of these agents.

OBJECTIVE To examine the use of piperacillin-tazobactam compared with cefepime in 90-day

mortality in patients treated empirically for sepsis, using instrumental variable analysis
of a 15-month piperacillin-tazobactam shortage.

DESIGN, SETTING, AND PARTICIPANTS In a retrospective cohort study, hospital admissions

at the University of Michigan from July 1, 2014, to December 31, 2018, including a
piperacillin-tazobactam shortage period from June 12, 2015, to September 18, 2016, were
examined. Adult patients with suspected sepsis treated with vancomycin and either
piperacillin-tazobactam or cefepime for conditions with presumed equipoise between
piperacillin-tazobactam and cefepime were included in the study. Data analysis was
conducted from December 17, 2022, to April 11, 2023.

MAIN OUTCOMES AND MEASURES The primary outcome was 90-day mortality. Secondary
outcomes included organ failure–free, ventilator-free, and vasopressor-free days. The
15-month piperacillin-tazobactam shortage period was used as an instrumental variable
for unmeasured confounding in antibiotic selection.

RESULTS Among 7569 patients (4174 men [55%]; median age, 63 [IQR 52-73] years) with
sepsis meeting study eligibility, 4523 were treated with vancomycin and piperacillin-
tazobactam and 3046 were treated with vancomycin and cefepime. Of patients who
received piperacillin-tazobactam, only 152 (3%) received it during the shortage. Treatment
groups did not differ significantly in age, Charlson Comorbidity Index score, Sequential Organ
Failure Assessment score, or time to antibiotic administration. In an instrumental variable
analysis, piperacillin-tazobactam was associated with an absolute mortality increase of 5.0%
at 90 days (95% CI, 1.9%-8.1%) and 2.1 (95% CI, 1.4-2.7) fewer organ failure–free days,
1.1 (95% CI, 0.57-1.62) fewer ventilator-free days, and 1.5 (95% CI, 1.01-2.01) fewer
vasopressor-free days.

CONCLUSIONS AND RELEVANCE Among patients with suspected sepsis and no clear indication
for antianaerobic coverage, administration of piperacillin-tazobactam was associated with
higher mortality and increased duration of organ dysfunction compared with cefepime.
These findings suggest that the widespread use of empirical antianaerobic antibiotics
in sepsis may be harmful.

Author Affiliations: Author

affiliations are listed at the end of this
article.
Corresponding Author: Rishi
Chanderraj, MD, MSc, Infectious
Diseases Section, Ann Arbor Veterans
Affairs Hospital, 2215 Fuller Rd,
JAMA Intern Med. 2024;184(7):769-777. doi:10.1001/jamainternmed.2024.0581 Room B804, Ann Arbor, MI 48105
Published online May 13, 2024. ([email protected]).

(Reprinted) 769
© 2024 American Medical Association. All rights reserved.
Downloaded from jamanetwork.com by Universidad de Chile user on 10/18/2024
 Research Original Investigation Mortality of Patients With Sepsis Administered Piperacillin-Tazobactam vs Cefepime

S
epsis, life-threatening organ dysfunction due to
infection,1 is the foremost cause of in-hospital mortal- Key Points
ity, unplanned readmission, and inpatient hospitaliza-
Question In patients with sepsis without a specific indication for
tion costs in the US.2-6 More than 65% of patients with sepsis antianaerobic antibiotics, is use of a combination of vancomycin
receive broad-spectrum antibiotics targeting methicillin- and piperacillin-tazobactam associated with increased mortality
resistant Staphylococcus aureus and Pseudomonas aeruginosa.7,8 risk compared with a combination of vancomycin and cefepime?
The most common empirical regimens are vancomycin in com-
Findings In this cohort study of 7569 patients with sepsis,
bination with either piperacillin-tazobactam or cefepime.7-12 an instrumental variable analysis using a 15-month
These 2 regimens are assumed to have equipoise, and guide- piperacillin-tazobactam shortage as a natural experiment
lines recommend either as empirical treatment for hospital- was conducted. The results found that treatment with vancomycin
associated pathogens.13-16 and piperacillin-tazobactam was associated with higher mortality
Yet unlike cefepime, piperacillin-tazobactam has potent than vancomycin combined with cefepime.
activity against anaerobic gut bacteria,17,18 which have a protec- Meaning The findings of this study suggest that, for patients
tive influence in numerous conditions.18-21 The clinical signifi- with sepsis not requiring antianaerobic antibiotics, a combination
cance of this difference remains uncertain.22,23 Multiple obser- of vancomycin and piperacillin-tazobactam may pose a greater
vational and experimental studies have suggested that among mortality risk than vancomycin and cefepime.
critically ill patients, combination therapy with piperacillin-
tazobactam and vancomycin is associated with increased
mortality compared with cefepime and vancomycin.18-20,24 identify eligible patients. Sepsis surveillance criteria were ad-
However, a recent pragmatic randomized clinical trial of cefe- justed to remove the 4-day antibiotic requirement,29 given the
pime and piperacillin-tazobactam among hospitalized patients focus on the initial 24-hour period. Thus, eligible patients met
(ACORN) found no significant difference in mortality across treat- the following criteria: (1) had blood samples drawn for cultures
ment groups.25 This study was limited by analyzing only short- upon arrival to the ED, (2) had evidence of acute organ dysfunc-
term outcomes (14 days), frequent treatment crossover, and tion in the first 24 hours after ED presentation, and (3) received
relatively low patient acuity. antibiotics for at least 1 day upon ED arrival. Among patients with
Given the importance of understanding the relative efficacy multiple admissions during the study period, we analyzed their
of the most prevalent antibiotic regimens for critically ill patients first hospital admission. We excluded patients transferred from
with sepsis, we used a nationwide shortage of piperacillin- outside hospitals, for whom reliable infection sources and
tazobactam to study differences in outcomes between these antibiotic administration data were unavailable.
regimens as empirical treatment for sepsis.26-28 This shortage To identify a population with equipoise between treatment
resulted in a substantial shift in antibiotic selection, which we groups, we excluded patients with indications for antianaerobic
leveraged to test the hypothesis that empirical therapy with antibiotic therapy identified within 24 hours of presentation. Spe-
piperacillin-tazobactam for sepsis is associated with increased cifically, we excluded patients with necrotizing, intra-abdominal,
90-day mortality compared with cefepime. or head and neck infections. Patients with central nervous sys-
tem infections were excluded, given cefepime’s superior blood-
brain barrier penetration. A combination of screening algorithms
of International Statistical Classification of Diseases and Related
Methods Health Problems, Tenth Revision claims data entered within the
Study Design and Data Source first 24 hours of presentation,7,31 medical record abstraction, mi-
This retrospective cohort study analyzed electronic health rec- crobiologic, and radiographic data was used to identify infectious
ord data from the University of Michigan Research Data Ware- sources and excluded diagnoses and conditions (Figure 1).
house. The University of Michigan Institutional Review Board
deemed the study exempt from needing consent under 45 CFR Treatment Variable and Patient Covariate Definitions
§46, category 4 (secondary use of identifiable data). This study The treatment variable was 1 or more days of piperacillin-
followed the Strengthening the Reporting of Observational tazobactam within the first 48 hours of admission. Potential
Studies in Epidemiology (STROBE) reporting guideline. confounding differences in treatment selection and outcomes
between patients treated with piperacillin-tazobactam and
Population cefepime were accounted for by adjusting for demographic char-
Eligible patients were adults aged 18 years and older who pre- acteristics (age, sex, and race), comorbid illness, severity of acute
sented to the emergency department (ED) at the University of illness within the first 24 hours, time from presentation to the
Michigan between July 1, 2014, and December 31, 2018, who met ED to first antibiotic administration, and concurrent metronida-
modified Centers for Disease Control and Prevention sepsis sur- zole use within the first 48 hours. Race was included as a covar-
veillance criteria29,30 and received empirical treatment for iate in the analysis in recognition of health disparities and the po-
sepsis with intravenous vancomycin and either intravenous tential for systematic bias in sepsis treatment. Metronidazole was
piperacillin-tazobactam or cefepime within the first 24 hours considered a covariate as it is often administered alongside
of presentation to the ED. To emulate the conditions of a ran- cefepime without indication and has potent antianaerobic
domized trial, we considered only data available to treatment activity. Preexisting comorbid illness was quantified with the
prescribers within the first 24 hours of ED presentation to Charlson Comorbidity Index,32,33 and the severity of illness was

770 JAMA Internal Medicine July 2024 Volume 184, Number 7 (Reprinted) jamainternalmedicine.com

© 2024 American Medical Association. All rights reserved.

Downloaded from jamanetwork.com by Universidad de Chile user on 10/18/2024
 Mortality of Patients With Sepsis Administered Piperacillin-Tazobactam vs Cefepime Original Investigation Research

Figure 1. Cohort Creation Diagram for the Study Figure 2. Change in Antibiotic Use During Piperacillin-Tazobactam Shortage
of Piperacillin-Tazobactam Shortage and Sepsis
Cefepime
28 627 Screening admissions (July 2014 - December 2018) Piperacillin-tazobactam
40
Shortage period
9794 Excluded
1297 Outside hospital transfers
8497 Repeat admissions 30

Daily antibiotic use, %

18 833 First admission via emergency department
20
8958 Excluded
6491 Vancomycin not administered
2467 Neither piperacillin-tazobactam
nor cefepime administered 10

9875 Administration of vancomycin and either

piperacillin-tazobactam or cefepime 0
2015 2016 2017 2018 2019
2306 Excluded Year
1583 Intra-abdominal infection
438 Necrotizing infection Percentage of daily antibiotic use among hospitalized patients at University of
24 Head and neck infection Michigan during piperacillin-tazobactam shortage (June 12, 2015, to September
22 Central nervous system infection 18, 2016) (calculated as number of doses of piperacillin-tazobactam
239 Multiple excluded diagnoses
administered in the hospital on a particular day divided by number of doses
of all antibiotics administered in the hospital on a particular day.)
7569 Clinical equipoise between piperacillin-tazobactam
and cefepime

outcome is mediated solely through the treatment variable and

Claims data, medical records abstraction, microbiology test results, and was indirectly assessed in 2 ways. First, we confirmed a lack of
radiology findings were used to identify a cohort with presumed equipoise
independent association between the shortage period and out-
between treatment groups.
comes in multivariable regression models after controlling for
the treatment variable.38,39 Second, we confirmed that the re-
characterized by the maximal Sequential Organ Failure Assess- siduals from the second-stage regression were not associated with
ment score in the first 24 hours of presentation.1,34 the instrument. Independence, or a lack of association between
the instrument and unobserved confounders, was indirectly
Outcome Measures evaluated by comparing the distribution of covariates within and
The primary outcome was all-cause mortality within 90 days of outside the shortage period with standardized differences.41-43
ED presentation. Patient data were linked to the Social Security A threshold of standardized difference less than 0.1 was set to
Death Index to identify postdischarge deaths. In line with pre- denote balance between treatment groups.42,43
vious trials of critical illness and sepsis, secondary outcomes were We used the Durbin-Wu-Hausman test to assess endoge-
organ failure–free days, determined by the number of days within neity. Endogeneity signifies that a treatment variable is asso-
the first 28 days post-ED arrival without receipt of vasopressors, ciated with unobserved confounders, introducing bias in the
mechanical ventilation, or dialysis.35-37 estimated outcome. The Durbin-Wu-Hausman test compares
coefficient estimates from standard regression with those ob-
Instrumental Variable tained from the instrumental variable model designed to miti-
An instrumental variable analysis was used to emulate a clini- gate unobserved confounding. A significant difference in the
cal trial with random treatment assignment.38-40 An instrumen- estimates indicates the presence of endogeneity. In such cases,
tal variable is an exogenous factor that influences treatment the instrumental variable model accounts for hidden biases and
assignment and simulates randomization but is not linked to provides a more reliable estimation.44-46
other patient characteristics or confounders.38-40 The chosen in- In an instrumental variable analysis, the outcome of the
strument was a period of piperacillin-tazobactam shortage at the treatment is designed to reflect its use in the marginal patient.47
University of Michigan Healthcare System from June 12, 2015, Marginal patients are those whose piperacillin-tazobactam
to September 18, 2016 (Figure 2).26,27 Patients admitted both treatment status was due solely to the timing of admission (ie,
before and after the shortage were included in the cohort to miti- during the shortage or not) and not on confounding factors,
gate any bias due to temporal changes in sepsis outcomes. such as comorbid illness, acuity of illness, or source of infec-
The shortage period was evaluated to confirm that it satis- tion. The effective sample size was calculated using the pro-
fied conditions required for instrumental variables: relevance, portion of compliers.47 In this context, the effective sample size
exclusion restriction, and independence38,39 (eMethods in estimates the number of marginal patients in the cohort.
Supplement 1). Relevance, a robust association between the
instrument and treatment, was evaluated via an F test, with a Statistical Analysis
threshold of F greater than 10 indicating satisfactory instrument Data analysis was conducted from December 17, 2022, to
strength.40 The exclusion restriction denotes that an instrument’s April 11, 2023. Associations across the shortage period and

jamainternalmedicine.com (Reprinted) JAMA Internal Medicine July 2024 Volume 184, Number 7 771

© 2024 American Medical Association. All rights reserved.

Downloaded from jamanetwork.com by Universidad de Chile user on 10/18/2024
 Research Original Investigation Mortality of Patients With Sepsis Administered Piperacillin-Tazobactam vs Cefepime

treatment groups were evaluated using χ2 tests and t tests.

Unadjusted analyses without covariates were performed Results
using logistic regression for 90-day mortality and linear
regression for organ failure–free, ventilator-free, and Among 28 627 acute care hospitalizations meeting modified
vasopressor-free days. To account for patient differences, the sepsis surveillance criteria within the study period, 8497 (30%)
association between piperacillin-tazobactam treatment and were repeated admissions and excluded, 6491 (23%) had no
90-day mortality and secondary outcomes was evaluated by vancomycin treatment, 2467 (9%) had no piperacillin-
logistic and linear regression models adjusted for patient tazobactam or cefepime treatment, and 2306 admissions (8%)
characteristics. had an excluded diagnosis identified within the first 24 hours
In the instrumental variable analyses, 2-stage least- of admission (Figure 1). With further exclusions, the final
squares regression was performed after adjusting for the pa- sample included 7569 patients (4174 men [55%]; 3395 women
tient characteristics included in the analysis in the first stage [45%]; median age, 63 [IQR 52-73] years); further patient char-
to account for their potential association with treatment se- acteristics are listed in the Table and eTables 1-3 in Supple-
lection, and again in the second stage to account for their ment 1. In unadjusted analysis, piperacillin-tazobactam was
potential association with outcomes.42,48 The adjusted out- associated with a 2.6% increase in 90-day mortality (P = .01)
comes from the instrumental variable analysis represent the (Table; eTable 4 in Supplement 1) and a decrease in 90-day
mean predicted difference in the probability of death at 90 days survival (Figure 3, Table; eTables 4-8 in Supplement 1).
or the difference in organ failure-free days, ventilator-free days, The piperacillin-tazobactam shortage period was a strong
and vasopressor-free days within the first 28 days following instrument, exhibiting a robust association with piperacillin-
admission. Adjusted absolute rates of 90-day mortality and tazobactam treatment (F1,7531 = 4355; P < .001) (eTable 9 in
absolute organ failure–free days were estimated using predic- Supplement 1) and had no independent association with the
tive margins. primary outcome (eTable 10 in Supplement 1). Of piperacillin-
We performed sensitivity analyses to determine the tazobactam-treated patients, 97% (4371 of 4523) were admit-
robustness of the findings. To further mitigate time-related ted outside the shortage period and 3% (152 of 4523) were ad-
confounding, we added the year of admission as a covariate mitted within the shortage period. The estimated effective
to the original model. To evaluate the exclusion restriction, sample size of marginal patients was 7062, or 93% of the total
we included other antibiotics with prescription rate differ- cohort. Covariates were well balanced during and outside the
ences exceeding a standardized mean difference of 0.1 in the shortage period (eTable 1 in Supplement 1), without a clear tem-
first 48 hours as covariates to the original model. To address poral trend in mortality (eTable 4, eFigure 1 in Supplement 1).
potential survivor bias, we included repeat admissions dur- Other antibiotics impacted by the shortage included metro-
ing the study period in the analysis, treating each admission nidazole, fluroquinolones, macrolides, and β-lactams with-
as a separate observation while clustering SEs with a sand- out Pseudomonas coverage (eTable 3, eFigure 2 in Supple-
wich estimator. We performed a subgroup analysis among ment 1). The Durban-Wu-Hausman test yielded a nonsignificant
patients with only 1 admission during the time frame to result (H = 0.49; P = .48;), indicating no statistical evidence of
address the impact of postindex admission antibiotics. To endogeneity in the treatment variable. Despite this result, we
ensure the results’ consistency with the modeling method, recognized the possibility of remaining unmeasured confound-
we repeated the primary analysis using a nonlinear 2-stage ing that the test might not have captured and proceeded with
residual inclusion model with a second-stage logistic regres- the instrumental variable analysis.
sion model. Confidence intervals for the 2-stage residual In the instrumental variable analysis, which controls for
inclusion model were based on 3000 nonparametric boot- unobserved differences in patient characteristics, piperacillin-
strap samples with replacement. tazobactam treatment was associated with higher 90-day
To evaluate specifically whether the antianaerobic activ- mortality than cefepime treatment (22.5% vs 17.5%; P = .002),
ity of antibiotics led to differences in outcomes and mitigate with an absolute increase in 90-day mortality of 5.0% (95% CI,
any selection bias introduced by metronidazole use, we per- 1.9%-8.1%) (Figure 4; eTable 11 in Supplement 1). Piperacillin-
formed an additional analysis in which we considered any tazobactam treatment was associated with 2.1 fewer organ
antianaerobic treatment (either piperacillin-tazobactam or met- failure–free days (95% CI, 1.4-2.7), 1.1 fewer ventilator-free days
ronidazole) as the primary exposure. In addition to the covar- (95% CI, 0.57-1.62), and 1.5 fewer vasopressor-free days (95%
iates of the principal analysis, the year of admission and the CI, 1.01-2.01) (eTable 12 in Supplement 1). Sensitivity analy-
source of infection were included as covariates in the model. ses revealed uniform results in estimating the association of
Relevance, independence, and exclusion restriction of the piperacillin-tazobactam with higher 90-day mortality, fewer
piperacillin-tazobactam shortage period as an instrumental vari- organ failure–free, ventilator-free, and vasopressor-free days
able for this analysis were evaluated as in the primary analy- (eTables 10-19 in Supplement 1). There was no association be-
sis. We also assessed the consistency of the results of this sen- tween the shortage period and the residuals of the second-
sitivity analysis by refitting the model, using a nonlinear 2-stage stage models for primary and secondary outcomes (F < 0.01;
residual inclusion model with second-stage logistic regres- P > .99 for all).
sion. All analyses were conducted in R, version 4.3.2 (R Foun- In a secondary analysis, we analyzed the association be-
dation for Statistical Computing). With 2-sided, unpaired t tests, tween treatment with any antibiotic with antianaerobic activ-
P values <.05 were considered statistically significant. ity and patient outcomes. The piperacillin-tazobactam shortage

772 JAMA Internal Medicine July 2024 Volume 184, Number 7 (Reprinted) jamainternalmedicine.com

© 2024 American Medical Association. All rights reserved.

Downloaded from jamanetwork.com by Universidad de Chile user on 10/18/2024
 Mortality of Patients With Sepsis Administered Piperacillin-Tazobactam vs Cefepime Original Investigation Research

Table. Patient Characteristics

Patients, No. (%)

Piperacillin- Standardized
Cefepime tazobactam mean
Characteristic (n = 3046) (n = 4523) P value difference
Age, median (IQR), y 63 (51-73) 64 (52-73) .21 0.030
Sex
Male 1590 (52.2) 2584 (57.1)
<.001 0.099
Female 1456 (47.8) 1939 (42.9)
Racial minority groupa 465 (15.3) 842 (18.6) <.001 0.088
SOFA, median (IQR) 5 (3-6) 5 (3-6) .02 0.054
Charlson Comorbidity Index score, 2 (0-4) 1 (0-4) .11 0.037
median (IQR)
Surgical service admission 247 (8.1) 430 (9.5) .03 0.049
ICU admission 925 (30) 1496 (33) .01 0.058
Coronary artery disease 416 (13.7) 528 (11.7) .01 0.060
Congestive heart failure 668 (22) 936 (20.7) .20 0.030
Peripheral vascular disease 235 (7.7) 318 (7) .26 0.026
History of stroke 169 (5.6) 232 (5.1) .43 0.019
Dementia 175 (5.8) 203 (4.5) .02 0.057
Chronic pulmonary disease 783 (25.7) 1007 (22.3) <.001 0.081
Connective tissue disease 163 (5.4) 195 (4.3) .04 0.049
Peptic ulcer disease 12 (0.4) 9 (0.2) .14 0.036
Diabetes 828 (27.2) 1179 (26.1) .37 0.021
Hemiplegia 78 (2.6) 90 (2) .11 0.038
Kidney failure 626 (20.6) 877 (19.4) .21 0.029
Cancer 602 (19.8) 805 (17.8) .03 0.051
Cirrhosis 93 (3.1) 132 (2.9) .73 0.008
Metastatic cancer 289 (9.5) 439 (9.7) .76 0.007
Abbreviations: ICU, intensive care
HIV 9 (0.3) 13 (0.3) .95 0.002 unit; SOFA, Sequential Organ Failure
Infectious source Assessment.
a
Bacteremia 44 (1.4) 83 (1.8) .18 0.031 Race was included (as recorded in
electronic health records) as a
Skin and soft tissue infection 127 (4.2) 173 (3.8) .45 0.018
covariate in the analysis in
Urinary tract infection 210 (6.9) 312 (6.9) .99 <0.001 recognition of health disparities and
Pneumonia 568 (18.6) 713 (15.8) <.001 0.076 the potential for systematic bias in
sepsis treatment. The included
Unclear 2097 (68.8) 3242 (71.7) .01 0.062
categories are American
Overall difference .01 .085 Indian/Alaska Native, Asian, Black or
Outcome African American, Native Hawaiian
or Other Pacific Islander, and more
90-d Mortality 579 (19) 975 (21.6) .01 0.063
than 1 race.

period maintained its role as a potent instrument in this analy- using 2-stage regression (95% CI, 3%-21%; P < .001)
sis (F1,7531 = 455; P < .001) (eTable 20 in Supplement 1) with- (eTable 22 in Supplement 1) and a 14% increase in mortality
out a direct association with the primary outcome (eTable 21 using a nonlinear 2-stage residual inclusion model (95% CI,
in Supplement 1). Covariates were generally well balanced be- 6%-22%; P < .001) (eTable 23 in Supplement 1). Antianaerobic
tween patients who did or did not receive antianaerobic anti- antibiotics were associated with fewer ventilator-free days
biotic treatment (eTable 2 in Supplement 1), but patients who (−0.71; 95% CI, −0.36 to −1.07 days) (eTable 24 in Supple-
received antianaerobic treatment had a higher acuity of ill- ment 1), vasopressor-free days (−0.38; 95% CI, −0.7 to −0.04
ness as measured by Sequential Organ Failure Assessment score days) (eTable 25 in Supplement 1), and organ failure–free
(mean, 5.3 [2.5] vs 4.6 [2.0]; P < .001). The proportion of com- days (−1.82; 95% CI, −1.35 to −2.28 days) (eTable 26 in
pliers decreased to 25%, leading to an estimated effective Supplement 1). There was no association between the short-
sample size of 473 marginal patients. The Durbin-Wu- age period and the residuals of the second stage model in any
Hausman test indicated no significant endogeneity in the of the secondary analyses (F < 0.01; P > .99 for all). Given the
treatment variable (H = 2.4; P = .10), but, as in the primary discordance between these findings and the recently pub-
analysis, we proceeded with an instrumental variable analysis lished ACORN trial, 25 in a post hoc analysis we analyzed
to account for unmeasured confounding. 14-day outcomes, finding no significant difference between
In this sensitivity analysis, antianaerobic antibiotic treat- the piperacillin-tazobactam and cefepime groups (eFigure 3,
ment was associated with a 12% increase in 90-day mortality eTables 27-31 in Supplement 1).

jamainternalmedicine.com (Reprinted) JAMA Internal Medicine July 2024 Volume 184, Number 7 773

© 2024 American Medical Association. All rights reserved.

Downloaded from jamanetwork.com by Universidad de Chile user on 10/18/2024
 Research Original Investigation Mortality of Patients With Sepsis Administered Piperacillin-Tazobactam vs Cefepime

Figure 3. Ninety-Day Survival for Primary Cohort

100
Cefepime
Piperacillin-tazobactam

75
Percent alive
100

95
50
Percent alive 90

85

25 80

0 25 50 75 90
Days Log-rank: P = .01
0
0 25 50 75 90
Days
Cefepime
No. at risk 3046 2666 2535 2486 2467
No. of events 0 380 511 560 579 Unadjusted analysis reveals a
Piperacillin-tazobactam significant difference in 90-day
No. at risk 4523 3921 3711 3609 3548 survival among treatment groups.
No. of events 0 602 812 914 975 The inset shows the same data
on an enlarged y-axis.

Figure 4. Primary Instrumental Variable and Sensitivity Analyses for 90-Day Mortality Among Adults
Hospitalized With Suspected Sepsis Treated With Either Piperacillin-Tazobactam or Cefepime

Absolute difference, % Favors Favors

Model (95% CI) piperacillin-tazobactam cefepime
Primary analysis 5.0 (1.9-8.2)
Sensitivity analyses
Nonlinear model 5.3 (2.4-8.2)
Including year of admission 8.6 (4.3-13.0)
Including anatomic site of infection 4.7 (1.6-7.9)
Including other antibiotics 5.4 (2.3-8.6)
Primary analysis and sensitivity
–10 –5 0 5 10 analyses reveal a uniform effect
Absolute difference in 90-d mortality, % of piperacillin-tazobactam.
Whiskers indicate 95% CIs.

clinical practice settings7,8 in which infection sources are

Discussion often unclear and broad-spectrum coverage is typical.
Our findings should be interpreted in the context of sev-
In this study, among patients with sepsis without an indica- eral studies. Our primary finding, a 5.0% increase in mortality
tion for antianaerobic antibiotic therapy, empirical vancomy- among piperacillin-tazobactam-treated patients, is congruent
cin and piperacillin-tazobactam treatment was associated with both with preclinical animal studies19,49 and with 2 recent large
a 5% increase in 90-day mortality compared with empirical observational cohort studies. A study of 3032 patients receiv-
vancomycin and cefepime. This finding was robust after con- ing mechanical ventilation found that early exposure to anti-
trolling for temporal trends, source of infection, and othe anaerobic antibiotics (largely piperacillin-tazobactam) was
r antibiotic treatments. We also found that exposure to any associated with decreased 30-day survival.19 This finding was
antianaerobic antibiotic (eg, metronidazole) was associated subsequently recapitulated in a study of 15 908 patients pre-
with worse outcomes. These findings suggest that broad- senting to the emergency department.20 Yet in contrast, the re-
spectrum antibiotics with antianaerobic activity, such as cently published ACORN study,25 a pragmatic randomized
piperacillin-tazobactam, may cause harm in patients without clinical trial of piperacillin-tazobactam and cefepime, found no
a clear indication. significant difference in mortality across treatment groups.
To date, comparative trials of piperacillin-tazobactam and Several key differences in study design could potentially ex-
cefepime have mostly been conducted in patients with clearly plain this difference. First, the ACORN trial only compared mor-
defined infectious syndromes,22,23 often without the coad- tality at 14 days, whereas our primary analysis was performed
ministration of other antibiotics.22,23 This has limited the gen- at 90 days. When we reanalyzed our study outcomes at 14 days,
eralizability of their findings to the more complex clinical sce- we found no significant difference across treatment groups, sug-
narios often encountered in practice, including sepsis, in which gesting that this time point may be premature for detecting mor-
empirical antibiotics must be administered prior to pathogen tality differences attributable to initial antibiotic selection in
identification. The present study, using a natural experi- sepsis. Second, as a pragmatic trial, ACORN had considerable
ment, explores the impact of antibiotic choices in a popula- crossover (18% of patients) and neither restricted nor adjusted
tion that closely mirrors the diversity and uncertainty seen in for coadministration of other antianaerobic antibiotics (eg,

774 JAMA Internal Medicine July 2024 Volume 184, Number 7 (Reprinted) jamainternalmedicine.com

© 2024 American Medical Association. All rights reserved.

Downloaded from jamanetwork.com by Universidad de Chile user on 10/18/2024
 Mortality of Patients With Sepsis Administered Piperacillin-Tazobactam vs Cefepime Original Investigation Research

metronidazole). Third, only 54% of the ACORN cohort had of the instrument from unmeasured confounders) cannot be
sepsis and only 8% were admitted to an intensive care unit, proven. Nonetheless, we found a good balance of covariates
whereas our cohort exclusively comprised patients with across treatment groups for observable characteristics, and the
sepsis who had a higher severity of illness. Durbin-Wu-Hausman test suggested that the cohort condi-
Metronidazole, a potent antianaerobic antibiotic, was more tions approximated those of a randomized experiment. The
frequently prescribed during the piperacillin-tazobactam short- findings of this single-center study may not be generalizable
age period. While metronidazole is often added empirically to other settings or populations with different demographic
without a clear clinical rationale, it was also more commonly characteristics, treatment practices, or antibiotic resistance pat-
administered to patients with a higher severity of illness, po- terns. However, the University of Michigan is an extensive,
tentially introducing bias into the primary analysis. To miti- diverse health care system that serves a heterogeneous
gate this, we adjusted our analysis for metronidazole expo- patient population, and the focus on a single institution
sure when comparing piperacillin-tazobactam and cefepime. allowed us to have comprehensive and consistent data,
We also conducted a secondary instrumental variable analy- enhancing the accuracy of our analysis.
sis testing the association between any antianaerobic antibi-
otic treatment and outcomes, finding an association. The
effect size for this secondary analysis is relevant to the select
patients who received any antianaerobic antibiotics because
Conclusions
they were admitted outside the shortage period. This was a A combination of vancomycin and piperacillin-tazobactam is
smaller subset of patients, limiting the precision of the the most frequently prescribed empirical regimen for sepsis,7-11
results. Nevertheless, the congruence with the primary analy- and the findings of this cohort study suggest that this regi-
sis adds further evidence that unwarranted antianaerobic men may contribute to 1 additional death per every 20 pa-
antibiotic administration is associated with harm. tients with sepsis treated, potentially translating into thou-
sands of additional fatalities each year. Clinicians should
Strengths and Limitations be aware of the potential risks associated with antianaerobic
This study leverages a large cohort in which piperacillin- treatment and consider alternatives when treating patients with
tazobactam and cefepime were used under conditions of sepsis without a clear indication for antianaerobic antibiotic
presumed equipoise, as patients did not have specific indica- therapy. While future studies should investigate the mecha-
tions for antianaerobic antibiotics. A unique natural experi- nisms by which anaerobe depletion worsens clinical out-
ment, prompted by a piperacillin-tazobactam shortage, pro- comes, we believe our study has more immediate clinical
vided an opportunity to address unmeasured confounding, implications. Awareness of the complexities and potential im-
enabling a direct comparison of the association between pacts of different antibiotic regimens should foster a broader
these antibiotics and 90-day mortality in a clinical practice understanding of why judicious antibiotic use is essential,
setting. not merely to prevent the spread and evolution of future
Despite this study’s strengths, it carries several inherent antimicrobial resistance, but to immediately improve out-
limitations. As a retrospective cohort study relying on elec- comes of septic patients.
tronic medical records, it is vulnerable to unobserved con- Our study reveals the potential risks associated with em-
founding. To mitigate this confounding, we used an instru- pirical piperacillin-tazobactam in patients with sepsis without
mental variable analysis to emulate the conditions of a a specific indication for antianaerobic therapy. These findings
randomized trial. However, this approach relies on several should prompt reconsideration and further study of the wide-
assumptions, the most fundamental of which (independence spread use of empirical antianaerobic antibiotics in sepsis.

ARTICLE INFORMATION Arbor Healthcare System, Ann Arbor, Michigan Statistical analysis: Chanderraj, Admon, He,
Accepted for Publication: February 11, 2024. (Prescott); Veterans Affairs Center for Clinical Dickson, Sjoding.
Management Research, Ann Arbor, Michigan Obtained funding: Admon, Dickson.
Published Online: May 13, 2024. (Prescott); Department of Microbiology and Administrative, technical, or material support:
doi:10.1001/jamainternmed.2024.0581 Immunology, University of Michigan Medical Admon, Admon, Prescott, Prescott, Dickson,
Author Affiliations: Division of Infectious Diseases, School, Ann Arbor (Dickson); Computational Dickson, Sjoding, Sjoding.
Department of Internal Medicine, University of Medicine and Bioinformatics, University of Supervision: Admon, Dickson, Sjoding.
Michigan Medical School, Ann Arbor (Chanderraj, Michigan Medical School, Ann Arbor (Sjoding). Conflict of Interest Disclosures: Dr Admon
Albin); Medicine Service, Infectious Diseases Author Contributions: Dr Chanderraj had full reported receiving grants from the National
Section, Veterans Affairs Ann Arbor Healthcare access to all of the data in the study and takes Institutes of Health National Heart, Lung, and Blood
System, Ann Arbor, Michigan (Chanderraj); Weil responsibility for the integrity of the data and the Institute (NHLBI) during the study. Dr Albin
Institute for Critical Care Research & Innovation, accuracy of the data analysis. Drs Dickson and reported receiving personal fees from Charles River
Ann Arbor, Michigan (Chanderraj, Admon, Albin, Sjoding contributed equally. Laboratories, Shiongi Pharmaceuticals, and the
Prescott, Dickson, Sjoding); Division of Pulmonary Concept and design: Chanderraj, Admon, Dickson, American College of Physicians, and grant funding
and Critical Care Medicine, Department of Internal Sjoding. from the Michigan Institute of Clinical & Health
Medicine, University of Michigan Medical School, Acquisition, analysis, or interpretation of data: Research outside the submitted work. Dr Prescott
Ann Arbor (Admon, He, Nuppnau, Prescott, All authors. reported receiving grants from the Agency for
Dickson, Sjoding); Institute for Healthcare Policy Drafting of the manuscript: Chanderraj, Dickson. Healthcare Research and Quality, Veterans Affairs
and Innovation, University of Michigan, Ann Arbor Critical review of the manuscript for important Health Services Research and Development, and
(Admon, Prescott, Sjoding); Veterans Affairs Ann intellectual content: All authors. the NHLBI; salary support from the Centers for

jamainternalmedicine.com (Reprinted) JAMA Internal Medicine July 2024 Volume 184, Number 7 775

© 2024 American Medical Association. All rights reserved.

Downloaded from jamanetwork.com by Universidad de Chile user on 10/18/2024
 Research Original Investigation Mortality of Patients With Sepsis Administered Piperacillin-Tazobactam vs Cefepime

Disease Control and Prevention and Blue Cross Blue 8. Rhee C, Kadri SS, Dekker JP, et al; CDC increase risk of adverse clinical outcomes. Eur
Shield of Michigan outside the submitted work; and Prevention Epicenters Program. Prevalence of Respir J. 2023;61(2):2200910. doi:10.1183/
service on the Surviving Sepsis Campaign antibiotic-resistant pathogens in culture-proven 13993003.00910-2022
Guidelines. Dr Dickson reported receiving grants sepsis and outcomes associated with inadequate 20. Kullberg RFJ, Schinkel M, Wiersinga WJ.
from the NHLBI. Dr Sjoding reported receiving and broad-spectrum empiric antibiotic use. JAMA Empiric anti-anaerobic antibiotics are associated
grants from the NLHBI during the conduct of the Netw Open. 2020;3(4):e202899. doi:10.1001/ with adverse clinical outcomes in emergency
study. jamanetworkopen.2020.2899 department patients. Eur Respir J. 2023;61(5):
Funding/Support: This work was supported by 9. Prescott HC, Seelye S, Wang XQ, et al. Temporal 2300413. doi:10.1183/13993003.00413-2023
NHLBI grant T32HL007749 (Dr Chanderraj), trends in antimicrobial prescribing during 21. Baggs J, Jernigan JA, Halpin AL, Epstein L,
National Institute of Aging grant 5P30AG024824 hospitalization for potential infection and sepsis. Hatfield KM, McDonald LC. Risk of subsequent
(Dr Chanderraj), NHLBI grant R01HL144599 (Dr JAMA Intern Med. 2022;182(8):805-813. sepsis within 90 days after a hospital stay by type
Dickson), NHLBI grant K24HL159247 (Dr Dickson), doi:10.1001/jamainternmed.2022.2291 of antibiotic exposure. Clin Infect Dis. 2018;66(7):
NHLBI grant R01HL158626 (Dr Sjoding), Agency for 10. Wayne MT, Seelye S, Molling D, et al. Temporal 1004-1012. doi:10.1093/cid/cix947
Healthcare Research and Quality grant trends and hospital variation in time-to-antibiotics
R01HS026725 (Dr Prescott), and is the result of 22. Leibovici L, Yahav D, Paul M. Meta-analysis of
among veterans hospitalized with sepsis. JAMA a possible signal of increased mortality associated
work supported with resources and use of facilities Netw Open. 2021;4(9):e2123950. doi:10.1001/
at the Ann Arbor VA Medical Center. with cefepime use. Clin Infect Dis. 2010;51(11):1350-
jamanetworkopen.2021.23950 1351. doi:10.1086/657247
Role of the Funder/Sponsor: The funding 11. Jones BE, Jones MM, Huttner B, et al. Trends
organizations had no role in the design and conduct 23. Yahav D, Paul M, Fraser A, Sarid N, Leibovici L.
in antibiotic use and nosocomial pathogens in Efficacy and safety of cefepime: a systematic
of the study; collection, management, analysis, and hospitalized veterans with pneumonia at 128
interpretation of the data; preparation, review, or review and meta-analysis. Lancet Infect Dis. 2007;7
medical centers, 2006-2010. Clin Infect Dis. 2015; (5):338-348. doi:10.1016/S1473-3099(07)70109-3
approval of the manuscript; and decision to submit 61(9):1403-1410. doi:10.1093/cid/civ629
the manuscript for publication. 24. Brumit JC, Cluck DB, Brown TP, Tharp JL. The
12. Magill SS, O’Leary E, Ray SM, et al; Emerging association between empiric antimicrobial therapy
Disclaimer: This article does not represent the Infections Program Hospital Prevalence Survey
views of the Department of Veterans Affairs or the and the risk of clinical failure in critically ill patients
Team. Antimicrobial use in US Hospitals: with aspiration pneumonia. J Clin Pharm Ther.
US government. comparison of results from emerging infections 2022;47(11):1820-1825. doi:10.1111/jcpt.13773
Data Sharing Statement: See Supplement 2. program prevalence surveys, 2015 and 2011. Clin
Infect Dis. 2021;72(10):1784-1792. doi:10.1093/cid/ 25. Qian ET, Casey JD, Wright A, et al; Vanderbilt
REFERENCES ciaa373 Center for Learning Healthcare and the Pragmatic
Critical Care Research Group. Cefepime vs
1. Singer M, Deutschman CS, Seymour CW, et al. 13. Hooton TM, Bradley SF, Cardenas DD, et al; piperacillin-tazobactam in adults hospitalized with
The Third International Consensus definitions for Infectious Diseases Society of America. Diagnosis, acute infection: the ACORN randomized clinical
sepsis and septic shock (Sepsis-3). JAMA. 2016;315 prevention, and treatment of catheter-associated trial. JAMA. 2023;330(16):1557-1567. doi:10.1001/
(8):801-810. doi:10.1001/jama.2016.0287 urinary tract infection in adults: 2009 international jama.2023.20583
2. Liang L, Moore B, Soni A. National inpatient clinical practice guidelines from the Infectious
Diseases Society of America. Clin Infect Dis. 2010; 26. Haiduc M, Patel M, Walsh TL, Moffa MA,
hospital costs: the most expensive conditions by Bremmer DN. Impact of a cefepime shortage on
payer, 2017. Healthcare Cost and Utilization Project 50(5):625-663. doi:10.1086/650482
dosing regimens and outcomes in hospitalized
Statistical Briefs: Agency for Healthcare Research 14. Kalil AC, Metersky ML, Klompas M, et al. adults with febrile neutropenia. J Oncol Pharm Pract.
and Quality; July 2020. Statistical brief 261. Management of adults with hospital-acquired and 2021;27(2):297-304. doi:10.1177/1078155220918638
Accessed March 25, 2024. . https://hcup-us.ahrq. ventilator-associated pneumonia: 2016 clinical
gov/reports/statbriefs/sb261-Most-Expensive- practice guidelines by the Infectious Diseases 27. Park SC, Gillis-Crouch GR, Cox HL, et al.
Hospital-Conditions-2017.jsp#:~:text=In%202017% Society of America and the American Thoracic Consecutive antibiotic shortages highlight
2C%20aggregate%20hospital% Society. Clin Infect Dis. 2016;63(5):e61-e111. discrepancies between microbiology and
20costs,myocardial%20infarction%2C%20and% doi:10.1093/cid/ciw353 prescribing practices for intra-abdominal infections.
20heart%20failure Antimicrob Agents Chemother. 2023;95(5):
15. Metlay JP, Waterer GW, Long AC, et al. e01980-20. doi:10.1128/AAC.01980-20
3. Prescott HC, Langa KM, Iwashyna TJ. Diagnosis and treatment of adults with
Readmission diagnoses after hospitalization for community-acquired pneumonia: an official clinical 28. Vail E, Gershengorn HB, Hua M, Walkey AJ,
severe sepsis and other acute medical conditions. practice guideline of the American Thoracic Society Rubenfeld G, Wunsch H. Association between US
JAMA. 2015;313(10):1055-1057. doi:10.1001/jama. and Infectious Diseases Society of America. Am J norepinephrine shortage and mortality among
2015.1410 Respir Crit Care Med. 2019;200(7):e45-e67. patients with septic shock. JAMA. 2017;317(14):
doi:10.1164/rccm.201908-1581ST 1433-1442. doi:10.1001/jama.2017.2841
4. Liu V, Escobar GJ, Greene JD, et al. Hospital
deaths in patients with sepsis from 2 independent 16. Stevens DL, Bisno AL, Chambers HF, et al; 29. Rhee C, Dantes R, Epstein L, et al; CDC
cohorts. JAMA. 2014;312(1):90-92. doi:10.1001/ Infectious Diseases Society of America. Practice Prevention Epicenter Program. Incidence and
jama.2014.5804 guidelines for the diagnosis and management of trends of sepsis in US hospitals using clinical vs
skin and soft tissue infections: 2014 update by the claims data, 2009-2014. JAMA. 2017;318(13):
5. Mayr FB, Talisa VB, Balakumar V, Chang CH, Fine 1241-1249. doi:10.1001/jama.2017.13836
M, Yende S. Proportion and cost of unplanned Infectious Diseases Society of America. Clin Infect Dis.
30-day readmissions after sepsis compared with 2014;59(2):e10-e52. doi:10.1093/cid/ciu296 30. Centers for Disease Control and Prevention.
other medical conditions. JAMA. 2017;317(5):530- 17. Pettigrew MM, Gent JF, Kong Y, et al. CDC hospital toolkit for adult sepsis surveillance.
531. doi:10.1001/jama.2016.20468 Gastrointestinal microbiota disruption and risk of May 2018. Accessed September 12, 2023. https://
colonization with carbapenem-resistant www.cdc.gov/sepsis/pdfs/Sepsis-Surveillance-
6. Rhee C, Jones TM, Hamad Y, et al; Centers for Toolkit-Aug-2018_508.pdf
Disease Control and Prevention (CDC) Prevention Pseudomonas aeruginosa in intensive care unit
Epicenters Program. Prevalence, underlying causes, patients. Clin Infect Dis. 2019;69(4):604-613. 31. Chua KP, Fischer MA, Linder JA.
and preventability of sepsis-associated mortality in doi:10.1093/cid/ciy936 Appropriateness of outpatient antibiotic
US acute care hospitals. JAMA Netw Open. 2019;2 18. Shono Y, Docampo MD, Peled JU, et al. prescribing among privately insured US patients:
(2):e187571. doi:10.1001/jamanetworkopen.2018.7571 Increased GVHD-related mortality with ICD-10-CM based cross sectional study. BMJ. 2019;
broad-spectrum antibiotic use after allogeneic 364:k5092. doi:10.1136/bmj.k5092
7. Tedijanto C, Nevers M, Samore MH, Lipsitch M.
Antibiotic use and presumptive pathogens in the hematopoietic stem cell transplantation in human 32. Charlson ME, Pompei P, Ales KL, MacKenzie CR.
Veterans Affairs healthcare system. Clin Infect Dis. patients and mice. Sci Transl Med. 2016;8(339): A new method of classifying prognostic
2022;74(1):105-112. doi:10.1093/cid/ciab170 339ra71. doi:10.1126/scitranslmed.aaf2311 comorbidity in longitudinal studies: development
19. Chanderraj R, Baker JM, Kay SG, et al. In and validation. J Chronic Dis. 1987;40(5):373-383.
critically ill patients, anti-anaerobic antibiotics doi:10.1016/0021-9681(87)90171-8

776 JAMA Internal Medicine July 2024 Volume 184, Number 7 (Reprinted) jamainternalmedicine.com

© 2024 American Medical Association. All rights reserved.

Downloaded from jamanetwork.com by Universidad de Chile user on 10/18/2024
 Mortality of Patients With Sepsis Administered Piperacillin-Tazobactam vs Cefepime Original Investigation Research

33. Quan H, Li B, Couris CM, et al. Updating and JMIR Res Protoc. 2022;11(5):e36261. doi:10.2196/ samples. Stat Med. 2009;28(25):3083-3107.
validating the Charlson Comorbidity Index and 36261 doi:10.1002/sim.3697
score for risk adjustment in hospital discharge 38. Newhouse JP, McClellan M. Econometrics in 44. Durbin J. Errors in variables. Int Stat Rev. 1954;
abstracts using data from 6 countries. Am J Epidemiol. outcomes research: the use of instrumental 22(1/3):23-32.
2011;173(6):676-682. doi:10.1093/aje/kwq433 variables. Annu Rev Public Health. 1998;19:17-34. 45. Wu DM. Alternative tests of independence
34. Jones AE, Trzeciak S, Kline JA. The Sequential doi:10.1146/annurev.publhealth.19.1.17 between stochastic regressors and disturbances.
Organ Failure Assessment score for predicting 39. Baiocchi M, Cheng J, Small DS. Instrumental Econometrica. 1973:733-750. doi:10.2307/1914093
outcome in patients with severe sepsis and variable methods for causal inference. Stat Med.
evidence of hypoperfusion at the time of 46. Hausman JA. Specification tests in
2014;33(13):2297-2340. doi:10.1002/sim.6128 econometrics. Econometrica. 1978:1251-1271.
emergency department presentation. Crit Care Med.
2009;37(5):1649-1654. doi:10.1097/CCM. 40. Rassen JA, Brookhart MA, Glynn RJ, Mittleman doi:10.2307/1913827
0b013e31819def97 MA, Schneeweiss S. Instrumental variables II: 47. Harris KM, Remler DK. Who is the marginal
instrumental variable application-in 25 variations, patient? understanding instrumental variables
35. Gong MN, Schenk L, Gajic O, et al. Early the physician prescribing preference generally was
intervention of patients at risk for acute respiratory estimates of treatment effects. Health Serv Res.
strong and reduced covariate imbalance. J Clin 1998;33(5, pt 1):1337-1360.
failure and prolonged mechanical ventilation with Epidemiol. 2009;62(12):1233-1241. doi:10.1016/
a checklist aimed at the prevention of organ failure: j.jclinepi.2008.12.006 48. Valley TS, Sjoding MW, Ryan AM, Iwashyna TJ,
protocol for a pragmatic stepped-wedged cluster Cooke CR. Association of intensive care unit
trial of PROOFCheck. BMJ Open. 2016;6(6):e011347. 41. Wooldridge JM. Multiplicative panel data admission with mortality among older patients with
doi:10.1136/bmjopen-2016-011347 models without the strict exogeneity assumption. pneumonia. JAMA. 2015;314(12):1272-1279.
Econ Theory. 1997;13(5):667-678. doi:10.1017/ doi:10.1001/jama.2015.11068
36. Estcourt LJ, Turgeon AF, McQuilten ZK, et al; S0266466600006125
Writing Committee for the REMAP-CAP 49. Schuijt TJ, Lankelma JM, Scicluna BP, et al. The
Investigators. Effect of convalescent plasma on 42. Valley TS, Iwashyna TJ, Cooke CR, et al. gut microbiota plays a protective role in the host
organ support-free days in critically ill patients with Intensive care use and mortality among patients defence against pneumococcal pneumonia. Gut.
COVID-19: a randomized clinical trial. JAMA. 2021; with ST elevation myocardial infarction: 2016;65(4):575-583. doi:10.1136/gutjnl-2015-309728
326(17):1690-1702. doi:10.1001/jama.2021.18178 retrospective cohort study. BMJ. 2019;365:l1927.
doi:10.1136/bmj.l1927
37. Adhikari NK, Pinto R, Day AG, et al; LOVIT
Investigators. Lessening Organ Dysfunction With 43. Austin PC. Balance diagnostics for comparing
Vitamin C (LOVIT) trial: statistical analysis plan. the distribution of baseline covariates between
treatment groups in propensity-score matched

jamainternalmedicine.com (Reprinted) JAMA Internal Medicine July 2024 Volume 184, Number 7 777

© 2024 American Medical Association. All rights reserved.

Downloaded from jamanetwork.com by Universidad de Chile user on 10/18/2024