Clinical Infectious Diseases

MAJOR ARTICLE

Preventing New Gram-negative Resistance Through

Beta-lactam De-escalation in Hospitalized Patients
With Sepsis: A Retrospective Cohort Study
Besu F. Teshome,1,2 Taehwan Park,3 Joel Arackal,2 Nicholas Hampton,4 Marin H. Kollef,5 and Scott T. Micek1,2
1
Department of Pharmacy Practice, University of Health Sciences and Pharmacy in St. Louis, St. Louis, Missouri, USA; 2Center for Health Outcomes Research and Education, University of Health
Sciences and Pharmacy in St. Louis, St. Louis, Missouri, USA; 3College of Pharmacy and Health Sciences, St. John’s University, Queens, New York, USA; 4Center for Clinical Excellence, BJC
Healthcare, St. Louis, Missouri, USA; and 5Division of Pulmonary and Critical Care Medicine, Washington University School of Medicine, St. Louis, Missouri, USA

Downloaded from https://academic.oup.com/cid/article/79/4/826/7688896 by guest on 30 October 2024

Background. Whether antibiotic de-escalation reduces the risk of subsequent antibiotic resistance is uncertain. We sought to
determine if beta-lactam (BL) antibiotic de-escalation is associated with decreased incidence of new Gram-negative resistance in
hospitalized patients with sepsis.
Methods. In a retrospective cohort study, patients with sepsis who were treated with at least 3 consecutive days of BL antibiotics, the
first 2 days of which were with a broad-spectrum BL agent defined as a spectrum score (SS) of ≥7 were enrolled. Patients were grouped
into three categories: (1) de-escalation of beta-lactam spectrum score (BLSS), (2) no change in BLSS, or (3) escalation of BLSS. The
primary outcome was the isolation of a new drug-resistant Gram-negative bacteria from a clinical culture within 60 days of cohort
entry. Fine-Gray proportional hazards regression modeling while accounting for in-hospital death as a competing risk was performed.
Findings. Six hundred forty-four patients of 7742 (8.3%) patients developed new gram-negative resistance. The mean time to
resistance was 23.7 days yielding an incidence rate of 1.85 (95% confidence interval [CI]: 1.71–2.00) per 1000 patient-days. The lowest
incidence rate was observed in the de-escalated group 1.42 (95% CI: 1.16–1.68) per 1000 patient-days. Statistically significant
reductions in the development of new gram-negative resistance were associated with BL de-escalation compared to no-change
(hazards ratio (HR) 0.59 [95% CI: .48–.73]).
Conclusions. De-escalation was associated with a decreased risk of new resistance development compared to no change. This
represents the largest study to date showing the utility of de-escalation in the prevention of antimicrobial resistance.
Keywords. antibiotic de-escalation; beta-lactam antibiotics; resistance; sepsis; spectrum score.

Antimicrobial resistance represents a significant threat to the spectrum of antibiotics and/or discontinuing antibiotics as
healthcare systems worldwide. Infections caused by resistant soon as possible [4–6]. Shorter courses of antibiotics have prov­
pathogens have been associated with increased mortality, hos­ en to be as effective as longer, traditional treatment durations in
pital length-of-stay, and cost of care [1, 2]. Urgently, strategies several randomized controlled trials, and observational studies
to limit the development of resistance are a high priority for have associated shorter durations of antibiotic exposure with a
their continued utilization [3]. decreased risk of resistance development [7–12]. In accordance
Antibiotic de-escalation (ADE) has been recommended for with clinical practice guideline recommendations, clinicians
nearly a quarter century as a fundamental stewardship approach routinely implement the practice of ADE. However, the reward
that attempts to balance the need for early administration of of reduced antibiotic resistance downstream is supported by
broad-spectrum antibiotics intended to increase the probability limited evidence [6]. Explanations for this can be attributed to
of an appropriate empiric regimen partnered with a plan to re­ study design limitations, notably, brief study periods, which
duce the risk of subsequent antibiotic resistance by narrowing yielded small sample sizes and inadequate durations of patient
follow-up to capture the emergence of resistance [13, 14]. In ad­
dition, and of critical importance, there has been a lack of con­
Received 07 March 2024; editorial decision 02 May 2024; published online 6 June 2024 sensus regarding the definition of ADE across studies [6].
Correspondence: S. T. Micek, Center for Health Outcomes Research and Education, St Louis
College of Pharmacy, University of Health Sciences and Pharmacy in St Louis, 1 Pharmacy Place,
Utilizing an antibiotic spectrum score (SS) can address this dis­
St Louis, MO 63110 ([email protected]). crepancy by creating an objective way of measuring the antimi­
Clinical Infectious Diseases® 2024;79(4):826–33 crobial spectrum of activity patients are exposed to [15, 16]. This
© The Author(s) 2024. Published by Oxford University Press on behalf of Infectious Diseases
Society of America. All rights reserved. For commercial re-use, please contact [email protected] study aims to utilize a novel, cumulative SS approach to evaluate
for reprints and translation rights for reprints. All other permissions can be obtained through our the exposure to BLs with varying degrees of activity spectrum on
RightsLink service via the Permissions link on the article page on our site—for further information
please contact [email protected].
the risk of isolation of gram-negative pathogens with new resis­
https://doi.org/10.1093/cid/ciae253 tance phenotypes in hospitalized patients. In addition, we

826 • CID 2024:79 (15 October) • Teshome et al

identify essential subgroups that could influence antibiotic re­ was categorized as escalation (Supplementary Appendix 2).
sistance in this population, most notably, antibiotic duration. The 10% buffer was incorporated to minimize the influence of
de-escalating or escalating antibiotics for only 1–2 days in the
METHODS overall BL exposure time. All BL therapy administered in the
Study Design and Patient Population hospital for all admissions during the 60-day follow-up period
This was a retrospective cohort study, conducted between was incorporated into the cumulative SS, whereas BL exposure
1 December 2010 and 31 December 2017, of patients with sep­ in the outpatient setting was not. Companion antibiotics such
sis who were treated at Barnes-Jewish Hospital, a 1300-bed as aminoglycosides or fluoroquinolones were not included in
academic hospital in St. Louis, Missouri. Data for this study the daily spectrum score calculation.
were obtained from the electronic health record (EHR) which
includes administrative, clinical, laboratory, and pharmacy Outcomes
data repositories. We have previously used these data sources The primary outcome of interest was the isolation of a new
for similar observational studies related to antibiotic exposure drug-resistant gram-negative bacteria from a clinical culture.
Microbiological data were captured from the EHRs of all

Downloaded from https://academic.oup.com/cid/article/79/4/826/7688896 by guest on 30 October 2024

[11, 15]. The Washington University and University of Health
Sciences and Pharmacy in St. Louis Institutional Review 12 hospitals in the BJC Healthcare system. This allowed for
Boards approved the study protocol. the capture of data for patients that were cared for in other hos­
All patients ≥18 years of age with a discharge diagnosis code pitals within the same system before the cohort entry-sepsis ad­
for sepsis, severe sepsis, or septic shock using the International mission and after discharge. Each patient’s microbiological
Classification of Diseases, Ninth and Tenth Revisions, Clinical history 90 days before cohort entry through the first three
Modification codes [17] who were administered at least three days after cohort entry served as the reference period for sub­
consecutive days of BL antibiotics were included. sequent comparison. The previous 90 days were chosen to en­
sure that any resistance identified during follow-up had not
Definitions and Follow-Up been previously found, and resistant pathogens identified with­
Cohort entry was defined as the initiation date of a pivotal in the first 3 days of cohort entry were censored, as resistance
antibiotic defined as any broad-spectrum BL with an antimi­ was likely not associated with the BL exposure from the cohort
crobial spectrum score (SS) of ≥7, adopted from Ilges et al entry hospitalization. New resistance was evaluated with re­
(Supplementary Appendix 1) [15]. In brief, the spectrum score spect to the specific pathogen(s) isolated for each patient be­
method assigns a numerical score (1 to 11) based on tween days 4 and 60 after cohort entry [11] and defined as
organism-BL antibiotic activity pairings on each calendar day meeting at least one of the following: (1) third generation ceph­
a patient is exposed to BL during the follow-up period, with a alosporin resistance, (2) carbapenem resistance, and/or (3)
broader spectrum agent awarded a higher score [18, 19]. multidrug resistance as previously defined by Magiorakos
Patients had to receive two consecutive days of the pivotal BL et al [20]. Our criteria for antimicrobial susceptibility testing
antibiotic and at least 1 more day of any BL antibiotic on day can be found in Supplementary Appendix 3.
3 after cohort entry to be included.
The maximum BLSS was documented for each day of BL ex­ Covariates
posure following cohort entry (eg, if a patient received both Baseline characteristics consisted of time-independent covari­
ampicillin-sulbactam [SS = 7] and meropenem [SS = 10] on ates collected at cohort entry, which was composed of patient
the same day, the higher BLSS was counted). Cohort entry had demographics including age, sex and race, comorbidities,
to occur during the first 10 days of the patient’s hospitalization. Charlson comorbidity index score, APACHE II score, and
Patients were grouped into 3 categories: (1) de-escalation of length of hospital stay before cohort entry. Cumulative vari­
BLSS, (2) no change in BLSS, or (3) escalation of BLSS, using ables after cohort entry were treated as time-dependent covar­
the cumulative SS based on the BL exposure during follow-up. iates and included daily exposure to all BLs as well as
To determine an individual patient grouping, the pivotal BLSS companion antibiotic agents or classes. All antibiotic exposures
on cohort entry (day 1) was multiplied by the total number of were calculated using start and stop orders from the EHR, and
days of BL exposure during the study period, yielding the daily doses of the antibiotics were not factored in the analysis.
patient-specific cumulative BLSS. The patients’ actual BLSS Other time-dependent covariates based on daily exposure in­
received during follow-up was calculated and compared to the cluded treatment in the intensive care unit (ICU), receipt of va­
expected BLSS. If the actual BLSS was equal to the expected sopressors, mechanical ventilation, central venous catheter,
BLSS ± 10%, then the patient was categorized as having no chan­ and urinary catheterization. Additionally, the total length of
ge in BLSS. If the actual BLSS was less than the expected BLSS by stay in the hospital as well as discharge to rehabilitation or
≥10%, the patient was categorized as de-escalation. If the actual skilled nursing facilities were also accounted for in our analyt­
BLSS was greater than the expected BLSS by ≥10%, the patient ical model.

β-lactam De-escalation to Prevent Resistance • CID 2024:79 (15 October) • 827

Statistical Analysis individual BLSS groups, are described in Tables 1 and 2.
Descriptive analyses were used to summarize patient demo­ Most patients had no change in antibiotic spectrum (n =
graphics and clinical variables. The Fine-Gray proportional 4802 [62.0%]), followed by de-escalation (n = 1578 [20.4%])
hazards regression model was used to assess the influence of de- and escalation (n = 1362 [17.6%]), respectively. The mean
escalation, no change, and escalation and covariates on the de­ age for all groups was 60.3 years. Most patients were male
velopment of new resistance while accounting for in-hospital (n = 4347 (56.2%)) and White (n = 5224 (70.6%)). Patients
death as a competing risk [21]. Hospital discharge was not con­ in the escalation group spent more days in an ICU and had
sidered a competitive risk as follow-up cultures obtained after more days of mechanical ventilation, vasopressor administra­
discharge at any of the 12 hospitals in the hospital system tion, central venous catheter, and urinary catheter insertions
were evaluated for the primary outcome. Each subject was fol­ (Table 2). The escalation group also had more days of expo­
lowed from the day after the index date until the development sure to BLs and to agents with only gram-positive activity
of new resistance or death, whichever occurred first. Subjects (eg, vancomycin) compared to the other 2 groups (Table 2).
who did not develop new resistance or die during follow-up All patients in the escalation group had a cohort entry spec­
were censored at 60 days following cohort entry. Prior to mod­ trum score of <10, whereas patients in the other 2 groups

Downloaded from https://academic.oup.com/cid/article/79/4/826/7688896 by guest on 30 October 2024

eling, all covariates were tested for collinearity using a variance had a mix of cohort entry spectrum scores >7 (Supplementary
inflation factor threshold of >10. Additionally, all covariates Appendix 6). Patients in the no change group had higher
were initially screened for their confounding effects by calculat­ exposure to pivotal antibiotics compared to patients in the de-
ing the percent difference between the crude and adjusted mea­ escalation and escalation groups (Supplementary Appendix 7)
sures of association for each. Details are provided in and patients across all groups had minimal exposure to com­
Supplementary Appendixes 4 and 5. Multivariable Fine-Gray panion antibiotics (Table 2, Supplementary Appendix 8 and 9).
models compared de-escalation to no change and de-escalation The mortality rate was highest in the escalation group (41.3%)
to escalation with output reported as recommended by Austin compared to the de-escalation (18.1%) and no change groups
and Fine [22]. (26.6%).
The heterogeneity of treatment effect was assessed in pre­
selected subgroups. The justification for the inclusion of the Outcomes
subgroup analyses was to evaluate whether the difference in Six hundred forty-four patients (8.3%) developed new
resistance emergence was signaled by differences in demo­ gram-negative resistance, with a mean time to resistance
graphics, severity of illness markers and processes of care, of 23.7 days (standard deviation 15.0 days), yielding an in­
as well as the duration of BL and concomitant antibiotic ex­ cidence rate of 1.85 (95% confidence interval [CI]: 1.71–
posure, all of which have all been linked in some way either 2.00) per 1000 patient-days. The lowest incidence rate was
directly or indirectly by clinicians’ greater use of antibiotics observed in the de-escalated group, 1.42 (95% CI: 1.16–
in sicker patients. The interaction between the BL treatment 1.68) per 1000 patient-days (Table 3). There were statisti­
group at the yes–no level in these subgroups was deter­ cally significant reductions in the development of new
mined using Fine-Gray proportional hazards, followed by gram-negative resistance associated with de-escalation com­
forest-plot visualization of the treatment effects in the pared to no-change (HR 0.59 [95% CI: .48–.73])
subgroups. (Figure 1A). The hazard plot for the comparison of de-
We also explored multiple sensitivity analyses; the first was escalation to escalation revealed the 2 curves intersected
with survivors only, and the second included subjects with at over the first 30 days, suggesting a violation of the propor­
least one follow-up culture collected more than three days after tional hazard assumption (Figure 1B). Calculation of the
cohort entry. The latter was performed to limit the potential for hazard ratio stratified by time, estimated reduced risk in
surveillance bias due to different rates of follow-up cultures be­ the de-escalation group during the 30–60-day period (HR
tween patients with resistance and those who were censored. 0.74 [95% CI: .46–1.16]), but the result was not significant.
Finally, we varied the cutoff percentage to classify groups to The most common pathogen group isolated with new resis­
5%, 15%, and 20% difference between expected and actual cu­ tance were Enterobacterales, specifically third generation ceph­
mulative BLSS. All analyses were performed using SAS software alosporin and multidrug resistance with no distinction between
version 9.4 (Cary, North Carolina, USA). groups of interest. Carbapenem and multidrug-resistant
Pseudomonas aeruginosa were also common, most notably in
escalation patients (Table 3).
RESULTS

Participants Heterogeneity of Treatment-Effect Analyses

A total of 7742 patients were included. Demographic and clin­ We found no evidence of a treatment effect for the develop­
ical characteristics of the overall cohort, as well as the 3 ment of new gram-negative resistance with de-escalation as

828 • CID 2024:79 (15 October) • Teshome et al

Table 1. Description of Baseline Characteristics at Cohort Entry Among the Beta-lactam Spectrum Score Groups

Overall De-escalation No Change Escalation

(n = 7742) (n = 1578) (n = 4802) (n = 1362)
Patient age (y), mean ± SD 60.3 ± 15.5 59.9 ± 15.7 60.6 ± 15.6 59.7 ± 15.1
Male, n (%)a 4347 (56.2%) 852 (54.0%) 2657 (55.4%) 838 (61.6%)
Race, n (%)a
White 5224 (70.6%) 1060 (70.0%) 3197 (69.6%) 967 (74.5%)
Black 1985 (26.8%) 416 (27.5%) 1279 (27.9%) 290 (22.3%)
Other 196 (2.7%) 39 (2.6%) 116 (2.5%) 41 (3.2%)
APACHE score, mean ± SD 16.6 ± 5.6 17.0 ± 5.3 16.3 ± 5.6 17.1 ± 5.7
Charlson comorbidity index score, mean ± SD 6.1 ± 3.2 5.9 ± 3.2 6.1 ± 3.3 6.2 ± 3.1
Comorbid conditions, n (%)
Coronary artery disease 1828 (23.6%) 396 (25.1%) 1079 (22.5%) 353 (25.9%)
Congestive heart failure 2824 (36.5%) 582 (36.9%) 1703 (35.5%) 539 (39.6%)
Peripheral vascular disease 1135 (14.7%) 217 (13.8%) 732 (15.2%) 186 (13.7%)

Downloaded from https://academic.oup.com/cid/article/79/4/826/7688896 by guest on 30 October 2024

Cerebrovascular disease 1170 (15.1%) 270 (17.1%) 674 (14.0%) 226 (16.6%)
Chronic obstructive pulmonary disease 2844 (36.7%) 528 (33.5%) 1797 (37.4%) 519 (38.1%)
Connective tissue disease 368 (4.8%) 79 (5.0%) 239 (5.0%) 50 (3.7%)
Cirrhosis 1846 (23.8%) 355 (22.5%) 1123 (23.4%) 368 (27.0%)
Diabetes 2884 (37.3%) 618 (39.2%) 1806 (37.6%) 460 (33.8%)
Chronic kidney disease 2666 (34.4%) 565 (35.8%) 1624 (33.8%) 477 (35.0%)
HIV 113 (1.5%) 21 (1.3%) 71 (1.5%) 21 (1.5%)
Leukemia, lymphoma, myeloma, myelodysplastic syndrome 1424 (18.4%) 186 (11.8%) 801 (16.7%) 437 (32.1%)
Other primary malignancy 1024 (13.2%) 190 (12.0%) 684 (14.2%) 150 (11.0%)
Hospital length of stay prior to cohort entry, d, median (IQR) 0.0 (0.00–1.00) 0.0 (0.0–1.0) 0.0 (0.0–1.0) 0.0 (0.0–2.0)
ICU at cohort entry, no. (%) 5390 (69.6) 1168 (74.0) 3365 (70.1) 857 (62.9)
Beta lactam exposure prior to cohort entry, d, median (IQR) 0.0 (0.0–0.0) 0.0 (0.0–0.0) 0.0 (0.0–0.0) 0.0 (0.0–0.0)
Abbreviations: HIV, human immunodeficiency virus; IQR, interquartile range; SD, standard deviation.
a
Includes missing data.

Table 2. Description of Cumulative Variables After Cohort Entry

Overall De-escalation No Change Escalation

(n = 7742) (n = 1578) (n = 4802) (n = 1362)
Follow-up after cohort entry, d, mean ± SD 44.9 ± 21.6 50.0 ± 18.3 44.3 ± 22.4 41.3 ± 21.2
Hospital, d, median (IQR) 13 (8–24) 15 (9–26.25) 11 (7–19) 22 (13–36)
ICU, d, median (IQR) 6 (3–13) 7 (3–14) 5 (2–11) 13 (6–24)
Mechanical ventilation, d, median (IQR) 2 (0–8) 2 (0–9) 1 (0–7) 6 (0–14)
Central venous catheter, d, median (IQR) 10 (5–20) 12 (6–23) 9 (5–17) 20 (10–33)
Urinary catheter, d, median (IQR) 8 (3–15) 9 (4–17) 6 (2–12) 12 (5–22)
Vasopressor, d, median (IQR) 3 (0–7) 3 (0–7) 3 (0–6) 6 (2–12)
Antibiotic exposure, d, median (IQR)
Beta lactam 9 (6–16) 12 (8–18) 8 (5–12) 16 (10–26)
Clindamycin 0 (0–0) 0 (0–0) 0 (0–0) 0 (0–0)
Colistin 0 (0–0) 0 (0–0) 0 (0–0) 0 (0–0)
Agents with gram-positive activity only 6 (3–11) 5 (3–10) 5 (3–9) 12 (7–19)
Macrolides 0 (0–0) 0 (0–0) 0 (0–0) 0 (0–3)
Metronidazole 0 (0–2) 0 (0–3) 0 (0–2) 0 (0–4)
Fluoroquinolones 0 (0–0) 0 (0–0) 0 (0–0) 0 (0–0)
Sulfamethoxazole-trimethoprim 0 (0–0) 0 (0–0) 0 (0–0) 0 (0–0)
Tetracyclines 0 (0–0) 0 (0–0) 0 (0–0) 0 (0–0)
Aminoglycosides 0 (0–0) 0 (0–0) 0 (0–0) 0 (0–1)
Abbreviations: ICU, intensive care unit; IQR, interquartile range; SD, standard deviation.

compared with no change in BLSS amongst any subgroups. Sensitivity Analyses

Increasing duration of BL exposure demonstrated a reduced The results of the sensitivity analyses were consistent with the
risk in the de-escalation group compared to the no change primary analysis. Amongst survivors only, there were statisti­
group, but the test for interaction was not significant (Figure 2). cally significant reductions in the development of new

β-lactam De-escalation to Prevent Resistance • CID 2024:79 (15 October) • 829

Table 3. Bacterial Pathogens that Developed New Resistance

Overall De-escalation No Change Escalation

(n = 644) (n = 112) (n = 431) (n = 101)
Incidence rate per 1000 person-d (95% CI) 1.85 (1.71–2.00) 1.42 (1.16–1.68) 2.03 (1.84–2.22) 1.80 (1.45–2.15)
Time to new resistance, d, mean ± SD 23.7 ± 15.0 26.0 ± 14.1 22.0 ± 15.1 28.4 ± 14.4
Pathogens
Acinetobacter baumannii complex
Carbapenem resistant 39 (6.0) 4 (3.6) 29 (6.7) 6 (5.9)
MDR 43 (6.7) 6 (5.4) 30 (7.0) 7 (6.9)
Enterobacterales
3rd generation cephalosporin resistance 329 (51.0) 63 (56.3) 229 (53.1) 37 (36.3)
Carbapenem resistant 51 (7.9) 7 (6.3) 36 (8.4) 8 (7.8)
MDR 251 (38.9) 41 (36.6) 175 (40.6) 35 (34.3)
Pseudomonas aeruginosa
Carbapenem resistant 155 (24.0) 23 (20.5) 92 (21.3) 40 (39.2)

Downloaded from https://academic.oup.com/cid/article/79/4/826/7688896 by guest on 30 October 2024

MDR 125 (19.4) 20 (17.9) 72 (16.7) 32 (31.4)
Other MDR non-fermenting GNR 29 (4.5) 7 (6.3) 18 (4.2) 4 (3.9)
Source of isolationa
Blood 94 (14.6) 13 (11.6) 70 (16.2) 11 (10.8)
Respiratory specimen 262 (40.6) 51 (45.5) 160 (37.1) 51 (50.0)
Urine 163 (25.3) 28 (25.0) 111 (25.8) 24 (23.5)
Other 133 (20.6) 22 (19.6) 94 (21.8) 17 (16.7)
Data are presented as number (%).
Abbreviations: GNR, Gram-negative rod; MDR, multidrug resistant.
a
Some pathogens were isolated from multiple sites in the same patient.

gram-negative resistance associated with de-escalation com­ First, we used cumulative spectrum scores to facilitate the classifi­
pared to no-change but not de-escalation compared to escala­ cation of each patient’s BL regimen during sepsis rather than fo­
tion between 30 and 60 days (Supplementary Appendix 10). cusing on individual BL agents. This allowed us to create a
Similarly, in patients with at least one follow-up culture, there consistent, objective definition of de-escalation and escalation,
were statistically significant reductions in the development of which has proven to be a challenge in most other studies [6].
new Gram-negative resistance associated with de-escalation Second, our designation of de-escalation, escalation, or no change
compared to no-change (Supplementary Appendix 11). The was based on a patient’s entire BL exposure profile following ini­
risk for new resistance when adjusting the buffer threshold for tiation of broad-spectrum therapy and not based only on the early,
the actual versus expected BLSS is reported in Supplementary initial phase of treatment as commonly reported (eg, comparing
Appendix 12, with results like the 10% difference used for the BLSS on day 2 vs day 4). Limiting the evaluation using a fixed
primary analysis. time point, early in the treatment course, could increase the risk
of misclassifying patients who received longer courses or multiple
antibiotic regimens during their admission. We feel our method­
DISCUSSION
ological strategies overcome these obstacles and allow us to de­
The practice of ADE is recommended by several clinical practice scribe the risk of resistance associated with an ADE approach,
guidelines and position statements as a safe approach to treat­ which has been historically difficult to predict. Finally, we evalu­
ment and the primary method of preventing the development ated BL exposure as a time-dependent variable to account for
of new resistance [3, 6, 23, 24]. Unfortunately, evidence linking the changeability that often occurs during the treatment of sepsis.
ADE to resistance prevention has been elusive. Our study sug­ This approach allowed for an accurate evaluation of the influence
gests that de-escalation is associated with a reduced risk of de­ of ADE on the development of resistance over the entire time each
veloping new resistance in clinically relevant cultures within individual patient had exposures to BLs [25].
60 days of BL initiation compared to no change in BL spectrum. Another interesting finding from our study, although from a
To our knowledge, this is the first and largest population-based subgroup and only hypothesis-generating, was the magnitude
retrospective study confirming that the practice of BL de- and significance of ADE on reducing the risk of gram-negative
escalation is associated with a reduction in the development of resistance the longer patients are exposed to BLs. This supports
new gram-negative pathogen resistance in a population of hospi­ our previous work, which found an association between each
talized patients with a discharge diagnosis of sepsis. Three princi­ additional day of exposure to antipseudomonal BLs and the de­
pal elements set this study apart from other similar evaluations. velopment of new resistance, a relationship that proved to be

830 • CID 2024:79 (15 October) • Teshome et al

Figure 1. A,B, Cumulative incidence of new-drug resistant gram-negative pathogen isolation. The cumulative incidence curves are plotted to the last event time in each Downloaded from https://academic.oup.com/cid/article/79/4/826/7688896 by guest on 30 October 2024
treatment group. The cumulative incidence of new-drug resistant gram-negative pathogen isolation over the course of 60 d was 1.42 per 1000 d (95% CI: 1.16–1.68) in the
de-escalation group (A and B—solid line), 2.03 per 1000 d (95% CI: 1.84–2.22) in the no change group (A—dashed line) and 1.80 per 1000 d (95% CI: 1.45–2.15) in the
escalation group B—dashed line). Abbreviation: CI, confidence interval.

linear with each additional day of exposure up to >21 days [11, escalation. De Bus et al observed that the cumulative incidence
26]. This finding also suggests that ADE may not be associated of the emergence of resistance to the initial BL antibiotic admin­
with a reduced risk of gram-negative resistance when shorter istered in an ICU population was not significantly different on
courses of BL therapy are provided. day 14 in patients classified as de-escalation compared to the
Our findings contrast with two previous retrospective cohort continuation of empirical treatment [13]. Likewise, Weiss et al
studies that did not use spectrum scores to define BL de- reported no difference between patients treated with a BL de-

β-lactam De-escalation to Prevent Resistance • CID 2024:79 (15 October) • 831

 Downloaded from https://academic.oup.com/cid/article/79/4/826/7688896 by guest on 30 October 2024
Figure 2. Heterogeneity of the treatment effect of time to new-drug resistant gram-negative pathogen isolation, De-escalation vs no change. A hazard ratio 1.0 indicates
an increased risk of resistance. Shown are the covariate-adjusted and model-based estimates of the treatment effect for selected subgroups. Abbreviations: CI, confidence
interval; HR, hazard ratio.

escalation strategy compared to no-de-escalation on the acqui­ Our study has several important limitations. First, our study
sition of all multidrug-resistant bacteria strains, ESBL only selected patients for cohort entry from a single academic
Enterobacteriaceae, resistant Pseudomonas aeruginosa, AmpC- medical center, which necessarily limits the generalizability of
hyperproducing Enterobacteriaceae and methicillin-resistant our findings. Second, the time frame for our study is 2010–
Staphylococcus aureus within 21 days of treatment initiation 2017 and may not represent the change in antibiotic utilization
[14]. A possible explanation for the lack of association between patterns that occurred during and after this coronavirus disease
ADE and a reduction in resistance was the shorter follow-up 2019 (COVID-19) pandemic. However, during a 7-year period
window than utilized in our study. The average time to resis­ of antibiotic use outside of a pandemic, prescribing patterns and
tance development in our study was 23.7 days (standard devia­ incidence rates of new resistance remained consistent year-over
tion 15.0 days). year (Supplementary Appendix 13). Third, our data set did not

832 • CID 2024:79 (15 October) • Teshome et al

include any possible antibiotic exposures or any relevant micro­ 3. Evans L, Rhodes A, Alhazzani W, et al. Surviving sepsis campaign: international
guidelines for management of sepsis and septic shock 2021. Crit Care Med 2021;
biologic data from hospitals outside our system that the patients 49:e1063–1143.
may have encountered during the study period. Fourth, our re­ 4. Kollef MH. Antimicrobial therapy of ventilator-associated pneumonia: how to se­
lect an appropriate drug regimen. Chest 1999; 115:8–11.
liance on ICD-9-CM and ICD-10-CM codes and the retrospec­ 5. Barlam TF, Cosgrove SE, Abbo LM, et al. Implementing an antibiotic stewardship
tive nature of our design may lead to misclassification bias and program: guidelines by the Infectious Diseases Society of America and the Society
for Healthcare Epidemiology of America. Clin Infect Dis 2016; 62:e51–77.
confounding. Fifth, to capture the study outcome we relied on
6. Tabah A, Bassetti M, Kollef MH, et al. Antimicrobial de-escalation in critically ill
non-standardized clinical cultures collected as part of the pa­ patients: a position statement from a task force of the European Society of
tient’s usual care for suspected or proven infection. To identify Intensive Care Medicine (ESICM) and European Society of Clinical
Microbiology and Infectious Diseases (ESCMID) Critically Ill Patients Study
earlier emergence of resistance, collection of fecal samples that Group (ESGCIP). Intensive Care Med 2020; 46:245–65.
detect changes in the gut microbiome is ideal. Finally, we did not 7. Chastre J, Wolff M, Fagon J-Y, et al. Comparison of 8 vs 15 days of antibiotic ther­
apy for ventilator-associated pneumonia in adults: a randomized trial. JAMA
account for many possible covariates including pharmacody­ 2003; 290:2588–98.
namic or pharmacokinetic differences between BLs, appropri­ 8. Uranga A, España PP, Bilbao A, et al. Duration of antibiotic treatment in
community-acquired pneumonia: a multicenter randomized clinical trial.
ateness of antibiotics, outbreaks of infections that could have JAMA Intern Med 2016; 176:1257–65.
occurred in various units or wards, or how changes to the micro­

Downloaded from https://academic.oup.com/cid/article/79/4/826/7688896 by guest on 30 October 2024

9. Yahav D, Franceschini E, Koppel F, et al. Seven versus 14 days of antibiotic ther­
apy for uncomplicated gram-negative bacteremia: a noninferiority randomized
biome could influence new, clinically relevant resistance.
controlled trial. Clin Infect Dis 2019; 69:1091–8.
In conclusion, ADE was associated with a reduced risk of 10. Bell BG, Schellevis F, Stobberingh E, Goossens H, Pringle M. A systematic review
new gram-negative resistance compared to no change in BL and meta-analysis of the effects of antibiotic consumption on antibiotic resis­
tance. BMC Infect Dis 2014; 14:13.
spectrum in patients hospitalized with sepsis. Clinicians should 11. Teshome BF, Vouri SM, Hampton N, Kollef MH, Micek ST. Duration of exposure
be vigilant in their efforts to de-escalate broad-spectrum BL to antipseudomonal β-lactam antibiotics in the critically ill and development of
new resistance. Pharmacotherapy 2019; 39:261–70.
therapy and limit exposure to the shortest effective duration 12. Yusuf E, Van Herendael B, Verbrugghe W, et al. Emergence of antimicrobial re­
to curb resistance emergence. Future studies are needed to con­ sistance to Pseudomonas aeruginosa in the intensive care unit: association with the
duration of antibiotic exposure and mode of administration. Ann Intensive Care
firm these findings and to rigorously evaluate ADE in other pa­ 2017; 7:72.
tient populations including organ transplant and other types of 13. De Bus L, Denys W, Catteeuw J, et al. Impact of de-escalation of beta-lactam an­
tibiotics on the emergence of antibiotic resistance in ICU patients: a retrospective
immunosuppressed patients.
observational study. Intensive Care Med 2016; 42:1029–39.
14. Weiss E, Zahar JR, Garrouste-Orgeas M, et al. De-escalation of pivotal beta-
Supplementary Data lactam in ventilator-associated pneumonia does not impact outcome and margin­
Supplementary materials are available at Clinical Infectious Diseases online. ally affects MDR acquisition. Intensive Care Med 2016; 42:2098–100.
Consisting of data provided by the authors to benefit the reader, the posted 15. Ilges D, Ritchie DJ, Krekel T, et al. Assessment of antibiotic de-escalation by spec­
trum score in patients with nosocomial pneumonia: a single-center, retrospective
materials are not copyedited and are the sole responsibility of the authors, so
cohort study. Open Forum Infect Dis 2021; 8:ofab508.
questions or comments should be addressed to the corresponding author. 16. Ilges D, Tande AJ, Stevens RW. A broad spectrum of possibilities: spectrum scores
as a unifying metric of antibiotic utilization. Clin Infect Dis 2023; 77:167–73.
Notes 17. Singer M, Deutschman CS, Seymour CW, et al. The third international consensus
definitions for sepsis and septic shock (sepsis-3). JAMA 2016; 315:801–10.
Author Contributions. S. T. M. had full access to all data in the study and
18. Gerber JS, Hersh AL, Kronman MP, Newland JG, Ross RK, Metjian TA.
takes responsibility for the integrity of the data and the accuracy of the data
Development and application of an antibiotic spectrum index for benchmarking
analysis. Concept and design: S. T. M., B. F. T., M. H. K., J. A. Acquisition, antibiotic selection patterns across hospitals. Infect Control Hosp Epidemiol
analysis, or interpretation of data: All authors. Drafting of the manuscript: 2017; 38:993–7.
S. T. M., M. H. K., B. F. T., T. P., J. A. Critical review of the manuscript 19. Madaras-Kelly K, Jones M, Remington R, Caplinger C, Huttner B, Samore M.
for important intellectual content: All authors. Statistical analysis: Description and validation of a spectrum score method to measure antimicrobial
T. P. Supervision: S. T. M., M. H. K. de-escalation in healthcare associated pneumonia from electronic medical re­
Acknowledgments. The authors recognize Abbey Jin, PharmD, Jaylana cords data. BMC Infect Dis 2015; 15:197.
Ahmetspahic, Seena Cherian, PharmD, Ashley Jose, PharmD, Diana 20. Magiorakos A-P, Srinivasan A, Carey RB, et al. Multidrug-resistant, extensively
drug-resistant and pandrug-resistant bacteria: an international expert proposal
Tran, PharmD, and Patrick Lapado, PharmD, for data collection support.
for interim standard definitions for acquired resistance. Clin Microbiol Infect
Potential conflicts of interest. M. H. K.’s efforts were supported by the
2012; 18:268–81.
Barnes-Jewish Hospital Foundation. B. F. T.’s contribution occurred while 21. Latouche A, Allignol A, Beyersmann J, Labopin M, Fine JP. A competing risks
he was working at the University of Health Sciences and Pharmacy in analysis should report results on all cause-specific hazards and cumulative inci­
St. Louis; he is now an employee of Melinta Therapeutics. All other authors dence functions. J Clin Epidemiol 2013; 66:648–53.
report no potential conflicts. 22. Austin PC, Fine JP. Practical recommendations for reporting Fine-Gray model
All authors have submitted the ICMJE Form for Disclosure of Potential analyses for competing risk data. Stat Med 2017; 36:4391–400.
Conflicts of Interest. Conflicts that the editors consider relevant to the con­ 23. Kalil AC, Metersky ML, Klompas M, et al. Management of adults with hospital-
tent of the manuscript have been disclosed. acquired and ventilator-associated pneumonia: 2016 clinical practice guidelines
by the Infectious Diseases Society of America and the American Thoracic
Society. Clin Infect Dis 2016; 63:e61–111.
References 24. Mazuski JE, Tessier JM, May AK, et al. The surgical infection society revised
1. Nelson RE, Hyun D, Jezek A, Samore MH. Mortality, length of stay, and health­ guidelines on the management of intra-abdominal infection. Surg Infect
care costs associated with multidrug-resistant bacterial infections among elderly (Larchmt) 2017; 18:1–76.
hospitalized patients in the United States. Clin Infect Dis 2022; 74:1070–80. 25. Munoz-Price LS, Frencken JF, Tarima S, Bonten M. Handling time-dependent
2. Nelson RE, Slayton RB, Stevens VW, et al. Attributable mortality of variables: antibiotics and antibiotic resistance. Clin Infect Dis 2016; 62:1558–63.
healthcare-associated infections due to multidrug-resistant gram-negative bacte­ 26. Teshome BF, Vouri SM, Hampton NB, Kollef MH, Micek ST. Evaluation of a ceiling
ria and methicillin-resistant Staphylococcus aureus. Infect Control Hosp effect on the association of new resistance development to antipseudomonal beta-
Epidemiol 2017; 38:848–56. lactam exposure in the critically ill. Infect Control Hosp Epidemiol 2020; 41:484–5.

β-lactam De-escalation to Prevent Resistance • CID 2024:79 (15 October) • 833