RESEARCH

Antimicrobial Resistance as Risk

Factor for Recurrent Bacteremia
after Staphylococcus aureus,
Escherichia coli, or Klebsiella spp.
Community-Onset Bacteremia
Salam Abbara, Didier Guillemot, David R.M. Smith, Salma El Oualydy,
Maeva Kos, Cécile Poret, Stéphane Breant, Christian Brun-Buisson, Laurence Watier

We investigated links between antimicrobial resistance

in community-onset bacteremia and 1-year bacteremia A ntimicrobial resistance (AMR) is a major global
health issue, associated with an estimated 4.95
million deaths worldwide in 2019 (1,2). Although the
recurrence by using the clinical data warehouse of Eu-
rope’s largest university hospital group in France. We effects of AMR on clinical and economic outcomes
included adult patients hospitalized with an incident have been studied extensively, relatively little is
community-onset Staphylococcus aureus, Escherichia known about the effects of AMR on infection recur-
coli, or Klebsiella spp. bacteremia during 2017–2019. rence, a significant event that results in substantial ill-
We assessed risk factors of 1-year recurrence using ness, death, and healthcare costs (3). Recurrence is of
Fine–Gray regression models. Of the 3,617 patients particular concern among bacteremia patients, who
included, 291 (8.0%) had >1 recurrence episode.
are often fragile and have underlying conditions,
Third-generation cephalosporin (3GC)-resistance was
because bacteremia is associated with high rates of
significantly associated with increased recurrence risk
after incident Klebsiella spp. (hazard ratio 3.91 [95% CI death and AMR (4). AMR in bacteremia is associated
2.32–6.59]) or E. coli (hazard ratio 2.35 [95% CI 1.50– with greater infection severity, higher risk for treat-
3.68]) bacteremia. Methicillin resistance in S. aureus ment failure, and longer length of hospital stay, all of
bacteremia had no effect on recurrence risk. Although which may affect risk for recurrence (5–7).
several underlying conditions and infection sources in- Few studies have investigated AMR as a poten-
creased recurrence risk, 3GC-resistant Klebsiella spp. tial risk factor for recurrent bacteremia, and all have
was associated with the greatest increase. These re- been limited to recurrence of infection attributable to
sults demonstrate a new facet to illness induced by the same bacterium that caused initial infection (8–13).
3GC-resistant Klebsiella spp. and E. coli in the com- Conversely, the few studies not targeting a specific
munity setting.
bacterial species or patient population (e.g., those with
underlying conditions) and studying risk factors asso-
Author affiliations: Institut National de la Santé et de la Recherche
ciated with recurrence within 1 year did not consider
Médicale, Montigny–Le-Bretonneux, France (S. Abbara,
AMR as a potential risk factor (14–16). However, when
D. Guillemot, C. Brun-Buisson, L. Watier); Versailles Saint Quentin
studying the link between AMR and recurrence, it is
en Yvelines University, Montigny–Le-Bretonneux (S. Abbara,
important to consider the prolonged microbial im-
D. Guillemot, C. Brun-Buisson, L. Watier); Institut Pasteur, Paris,
balance that broad-spectrum antibiotic exposure (i.e.,
France (S. Abbara, D. Guillemot, C. Brun-Buisson, L. Watier);
standard bacteremia treatment) can induce on the host
Paris–Cité University, Paris (S. Abbara, D. Guillemot,
microbiome. This imbalance includes ensuing effects
C. Brun-Buisson, L. Watier); Paris–Saclay University,
on host susceptibility to colonization and infection
Le Kremlin–Bicêtre, France (D. Guillemot); Assistance Publique–
(17) and effects on selection and duration of carriage
Hôpitaux de Paris, Paris (D. Guillemot, C. Poret, S. Breant);
of antibiotic-resistant bacteria, which, for instance,
University of Oxford, Oxford, UK (D.R.M. Smith); Plateforme de
can exceed 1 year for extended-spectrum β-lactamase
Données de Santé, Paris (S. El Oualydy, M. Kos)
(ESBL)–producing Enterobacteriaceae (18). AMR in an
DOI: https://doi.org/10.3201/eid3005.231555 initial bacteremia episode may thus increase risk for

974 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 30, No. 5, May 2024
 Antimicrobial Resistance after Bacteremia

recurrence attributable not only to the same bacterium important episode of community-onset, monomi-
that caused the initial infection but to any bacterium, crobial bacteremia attributable to S. aureus, E. coli, or
whether acquired in the community or hospital. As Klebsiella spp. in 14 AP-HP university hospitals dur-
such, studying bacteremia recurrence across all bac- ing January 1, 2017–December 31, 2019. We catego-
terial species and sources of infection seems clinically rized bacteremia episodes as community-onset if first
relevant. Moreover, it is particularly important to fo- positive blood culture was collected within 48 hours
cus on community-onset infections, given the increas- of admission; otherwise, we categorized the bactere-
ing spread of ESBL-producing Enterobacteriaceae in mia as hospital-acquired. We identified incident stays
community settings globally (19). by excluding stays by patients with any history of
In this study, we investigated the effect of AMR bacteremia within the previous 12 months, regard-
in incident community-onset bacteremia on the prob- less of microbial etiology and location of onset (i.e.,
ability of bacteremia recurrence within 1 year. We whether community-onset or hospital-acquired). We
restricted incident bacteremia episodes to infections excluded stays ending with death. To avoid including
caused by S. aureus, E. coli, and Klebsiella spp., 3 lead- early relapses, we defined recurrence as any clinically
ing pathogens responsible for community-onset bac- important episode of bacteremia (whatever the spe-
teremia, and to their leading forms of AMR of major cies, wherever the onset) occurring 7–365 days after
public health concern: methicillin resistance for S. au- hospital discharge from the incident episode (21). We
reus and third-generation cephalosporin (3GC) resis- identified bacteremia episodes by using microbio-
tance for E. coli and Klebsiella spp., the major mecha- logic results (i.e., positive blood cultures) and defined
nism of which is ESBL production (20). them as previously described (4). For statistical analy-
sis, we considered 2 main patient groups: those with
Material and Methods recurrence and those without.

Setting Data Collected

This observational study used routinely collected For each patient, data collected were sex, age, and
data extracted retrospectively from the clinical data date of death (if applicable). For each incident stay,
warehouse of the Assistance Publique–Hôpitaux de data collected were dates of admission and discharge;
Paris (AP-HP) (https://www.aphp.fr). AP-HP is the hospital care pathways (e.g., surgery, admission to
largest university hospital group in Europe, with 39 intensive care unit [ICU], and presence of a septic
hospitals mainly located in the Greater Paris area and shock); codes from the International Classification
totaling 1.5 million hospitalizations per year (10% of of Diseases, 10th Revision (ICD-10), for underlying
all hospitalizations in France). We focused on 14 AP- conditions; and microbiologic results (e.g., types and
HP hospitals with acute care activity, covering ≈22% dates of microbiologic samples drawn, bacterial spe-
of all short stays in Île-de-France, the largest region cies isolated, and antibiotic-susceptibility results).
in France. The construction of the database and the Bacterial antibiotic susceptibilities were determined
included variables have been previously described by the laboratories of participating hospitals by using
(4). Available data include medical-administrative clinical breakpoints from Comité de l’Antibiogramme
data describing patient characteristics and hospital de la Société Francaise de Microbiologie–European
stays, as well as microbiological data including in- Committee on Antimicrobial Susceptibility Testing
fection etiology and antibiotic-susceptibility results. (22) and the qualitative susceptibility categories of
We obtained approval for data collection from the susceptible, standard dosing regimen (S), susceptible,
Scientific and Ethical Committee of the Assistance increased exposure (I), and resistant (R). For antibiot-
Publique–Hôpitaux de Paris on March 28, 2019. The ics of interest, we considered strains reported as I to
AP-HP clinical data warehouse initiative ensures that be resistant. When available, empirical therapy data
patient information and informed consent regarding were collected 24 hours before and after the collection
the different approved studies are in accordance with date of the first positive blood culture. For each recur-
European regulations on data protection and authori- rent stay, we collected only dates of hospital admis-
zation number 1980120 from the French Data Protec- sion and discharge and microbiologic results.
tion Authority.
Variables Studied
Study Population Patient variables included sex, age group, Charlson
The study population included all patients >18 years comorbidity index (calculated by using comorbidity-
of age who were hospitalized with a first clinically associated ICD-10 codes), and comorbidities (i.e.,

Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 30, No. 5, May 2024 975
RESEARCH

underlying conditions) defined with ICD-10 codes. on the adequacy of empirical treatment. For compa-
For each incident stay, other variables considered rability purposes, we estimated a multivariable logis-
include patient length of stay (LOS) with bacteremia tic regression model, considering the same covariates
(the number of days in hospital from the first positive as in the final Fine–Gray regression model. Adjusted
blood culture to the end of the stay), surgery, ICU ad- odds ratios calculated in the logistic regression model
mission, presence of septic shock, identified bacterial quantify associations between included variables and
species (S. aureus, E. coli, or Klebsiella spp.), antibiotic the odds of bacteremia recurrence, without consid-
susceptibility results, and infection sources (defined ering death as a competing event. Finally, to ensure
according to ICD-10 codes, as previously described) that the type of recurrence, to the same species or to
(4). Because the effect of resistance may differ ac- a different species, was not a confounding factor, we
cording to bacterial species, we considered a 6-class estimated a specific Fine–Gray regression model and
bacteria resistance composite variable (methicillin- a multivariable logistic regression model in each of
susceptible S. aureus [MSSA] and methicillin-resistant the 2 subgroups, following the same methodology as
S. aureus [MRSA], 3GC-susceptible and resistant E. for the overall sample.
coli, and 3GC-susceptible and resistant Klebsiella spp.) We used HiveQL (https://hive.apache.org), Py-
We considered empirical therapy appropriate if >1 thon 3 (https://www.python.org), PySpark 2.4.3
antibiotic administered within 24 hours of drawing (https://spark.apache.org/docs/2.4.3), and R 4.0.0
the first positive blood culture was effective in vitro (The R Foundation for Statistical Computing, https://
on the isolated bacteria. cran.r-project.org) to perform the statistical analyses,
and we used the survival R package to compute Fine–
Statistical Analysis Gray regression models (24). This study follows the
We described all included patient and hospital stay Strengthening the Reporting of Observational Studies
characteristics according to the presence or absence in Epidemiology reporting guideline (25).
of recurrence. We also briefly described first recurrent
bacteremia and their etiologies. We used Fine–Gray Results
regression models to identify risk factors for recur- During 2017–2019, we identified 4,400 patients hospi-
rence within 1 year after an incident stay, considering talized with a community-onset bacteremia attribut-
death as a competing event. We calculated subdis- able to S. aureus, E. coli, or Klebsiella spp. We retained
tribution hazard ratios (HRs) by using Gray’s test of their first hospital stay with bacteremia. Of those first
the subdistribution function for univariate analyses stays, 6.9% (n = 304) were excluded because of history
and Fine–Gray regression models for multivariable of bacteremia within the previous 12 months (Figure
analyses. HRs represent the relative change in the 1). Among the remaining 4,096 patients, 11.7% (n =
instantaneous rate of the occurrence of recurrence in 479) died during their hospital stay and were exclud-
patients who are recurrence-free or who have experi- ed. In total, we included in the study 3,617 patients
enced death, considering patients who have died as with incident hospital stays with community-onset
nonexposed to recurrence (23). The direction of HRs bacteremia attributable to S. aureus, E. coli, or Klebsi-
also describes the direction of the effect of covariates ella spp. Of those, 8.0% (n = 291) sought treatment for
on the probability of recurrence occurring over time >1 recurrent bacteremia during the following year.
(incidence) (23). We considered variables that had a
p value <0.20 in univariate analyses in the multivari- Descriptive Analyses
able models. We selected variables in the multivari-
able models by using both backward and forward Incident Stays
stepwise methods, and we used a 2-tailed p value Patients with recurrence were more often male than
<0.05 to define statistical significance. We assessed patients without recurrence (56.7% male vs. 47.9%
proportional hazards assumptions in Fine–Gray re- female; p = 0.004) and were more likely to be <80
gressions. We forced the variables age and sex in the years of age (81.8% <80 vs. 73.4% >80; age distribu-
multivariable model because they usually are associ- tion p = 0.0005) (Table 1). Patients with recurrence
ated with bacterial infections. To confirm results and also had more underlying conditions (27% had a
because the effect of resistance may differ according to null Charlson comorbidity index vs. 44% of those
bacterial species, we conducted an analysis stratified without recurrence) and were almost twice as like-
by bacteria. To assess whether empirical treatment ly as their counterparts to have cancer (37.6% vs.
is a confounding factor, we performed an additional 19.5%; p<0.0001), renal disease (22.3% vs. 13.6%; p =
analysis considering only patients with information 0.002), or liver disease (13.8% vs. 7.5%; p = 0.0007).

976 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 30, No. 5, May 2024
 Antimicrobial Resistance after Bacteremia

Figure 1. Identification of incident stays of community-onset

bacteremia attributable to Staphylococcus aureus, Escherichia
coli, or Klebsiella spp., Assistance Publique–Hôpitaux de Paris
university hospital group, Paris, France, 2017–2019.

We observed no statistical difference (p>0.05) be- to the same species, >80% of isolates had the same
tween patients with and without recurrence in terms resistance phenotype as identified in the incident
of incident stay characteristics, including LOS with stay (Appendix Table 2).
bacteremia (median 7–8 days), rates of surgery, ICU
admission, and occurrence of septic shock. Howev- Regression Models
er, we observed significant differences regarding the Variables not selected for inclusion in the multivari-
infection source, isolated bacteria, and rates of AMR able analysis were heart failure, diabetes, systemic
(p<0.0001 for all). Compared with patients that did disease, LOS with bacteremia, surgery, ICU admis-
not have recurrence, recurrence was more often as- sion, and presence of septic shock (Table 2). We did
sociated with bacteremia without an identified in- not include Charlson scores in the multivariable
fection source (24.1% vs. 14.0%) or associated with a model because individual underlying conditions
digestive (12.1% vs. 9.5%) or device-related infection were preferred. In the final model, vascular disease,
(7.8% vs. 4.4%) and less often with urinary-source chronic lung disease, dementia, and presence of pa-
bacteremia (26.2% vs. 35.7%). Moreover, recurrences ralysis were not retained, and proportional hazards
were more often associated with incident infections assumptions were validated (p = 0.39).
attributable to 3GC-resistant E. coli (13.1% vs 7.6%),
3GC-susceptible (13.1% vs. 9.3%) or resistant (7.2% Underlying Conditions and Infection Sources
vs. 1.9%) Klebsiella spp. Certain infectious sources were associated with in-
creased recurrence risk within 1 year: absence of an
Recurrent Stays identified infection source (HR 2.26 [95% CI 1.60–
Patients with recurrence had an average of 2.3 3.19]), device-related infection (HR 1.93 [95% CI 1.16–
(range 2–8 stays) hospital stays with bacteremia 3.23]), and digestive tract infection (HR 1.57 [95% CI
over the study period, including their incident stay. 1.03–2.38]). Certain underlying conditions also were
First recurrent stays occurred within a median of identified as associated with increased recurrence risk
80 days (first quartile–third quartile 30.0–175.0 within 1 year: cancer (HR 2.03 [95% CI 1.58–2.62]), re-
after the incident stay) and were predominantly nal disease (HR 1.72 [95% CI 1.28–2.31]), and liver dis-
community-onset (n = 166/291 [57.0%]); median ease (HR 1.66 [95% CI 1.17–2.35]) (Table 2).
LOS was 11 days (first quartile–third quartile 6.0–
19.0 days). The most identified bacteria in first re- Antimicrobial Resistance
current stays were E. coli, Klebsiella spp., polymicro- Isolation of MRSA in incident bacteremia episodes
bial infection, S. aureus, and Pseudomonas aeruginosa did not affect the incidence of recurrence (HR 0.79
(Appendix Table 1, https://wwwnc.cdc.gov/EID/ [95% CI 0.29–2.19]; referent MSSA). Conversely,
article/30/5/23-1555-App1.pdf). In 47.4% of first isolation of 3GC-resistant E. coli (HR 2.02 [95% CI
recurrent episodes (n = 138/291), the same bacterial 1.41–2.91]; referent 3GC-susceptible E. coli) or 3GC-
species was identified as in the incident stay. This resistant Klebsiella spp. (HR 2.77 [95% CI 1.60–4.79];
rate was higher for E. coli (n = 92/174 [53%]) than referent 3GC-susceptible Klebsiella spp.) were associ-
for Klebsiella spp. (n = 25/59 [42%]) or S. aureus (n ated with an increased risk for recurrence (Table 3).
= 21/58 [36%]). In cases of recurrence attributable Cumulative incidence function curves of recurrence

Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 30, No. 5, May 2024 977
RESEARCH

over time (Figure 2) show the differential effect of tion on empirical treatment showed comparable re-
bacteria-resistance pairs on risk for recurrence, which sults, with a higher HR for 3GC-resistant Klebsiella
was highest for 3GC-resistant Klebsiella spp. Those spp. and no effect of adequacy of empirical treatment
results were similar in an analysis stratified by bacte- on recurrence risk (Appendix Table 4).
rial species (Table 4) and in the multivariable logistic Fine–Gray models limited to recurrence to the
regression model (Appendix Table 3). A sensitivity same or a different species (Appendix Tables 5–7)
analysis considering only stays (36%) with informa- showed similar HRs for 3GC-resistant Klebsiella spp.

Table 1. Characteristics of patients and their incident stays with community-onset bacteremia attributable to Staphylococcus aureus,
Escherichia coli, or Klebsiella spp., with and without recurrence, Assistance Publique–Hôpitaux de Paris university hospital group,
Paris, France, 2017–2019*
Characteristic With recurrence, n = 291 Without recurrence, n = 3,326 p value
Patients
Sex 0.004
M 165 (56.7) 1,594 (47.9)
F 126 (43.3) 1,732 (52.1)
Age group, y 0.0005
18–35 10 (3.4) 227 (6.8)
35–50 34 (11.7) 378 (11.4)
50–65 84 (28.9) 765 (23.0)
65–80 110 (37.8) 1,070 (32.2)
>80 53 (18.2) 886 (26.6)
Charlson comorbidity index† <0.0001
0 75 (26.6) 1,389 (44.1)
1–2 109 (38.7) 1,004 (31.9)
>2 98 (34.7) 7,59 (24.0)
Underlying conditions†
Cancer 106 (37.6) 615 (19.5) <0.0001
Heart failure 35 (12.4) 423 (13.4) 0.61
Diabetes 63 (22.3) 730 (23.2) 0.79
Vascular disease 23 (8.2) 340 (10.8) 0.18
Renal disease 63 (22.3) 430 (13.6) 0.002
Liver disease 39 (13.8) 237 (7.5) 0.0007
Chronic pulmonary disease 12 (4.3) 189 (6.0) 0.24
Dementia 9 (3.2) 178 (5.7) 0.06
Paralysis (hemiplegia or paraplegia) 4 (1.4) 84 (2.7) 0.21
Systemic disease 5 (1.8) 32 (1.0) 0.30
Incident stays
Length of stay with bacteremia, days
Median (first quartile–third quartile) 8.0 (4.0–15.5) 7.0 (3.0–15.0)
Duration, d 0.30
<7 139 (47.8) 1,730 (52.0)
7–14 74 (25.4) 706 (21.2)
14–30 50 (17.2) 613 (18.4)
>30 28 (9.0) 277 (8.3)
Surgery 37 (12.7) 426 (12.8) 0.97
ICU admission 70 (24.1) 718 (21.6) 0.30
Septic shock† 27 (9.6) 279 (8.9) 0.68
Infection source† <0.0001
None identified 68 (24.1) 442 (14.0)
Multiple sites 59 (20.9) 747 (23.7)
Urinary tract 74 (26.2) 1,125 (35.7)
Lower respiratory tract 14 (5.0) 190 (6.0)
Digestive tract 34 (12.1) 300 (9.5)
Device-related 22 (7.8) 137 (4.4)
Other 11 (3.9) 211 (6.7)
Bacteria resistance <0.0001
MSSA 54 (18.6) 737 (22.2)
MRSA 4 (1.4) 75 (2.2)
3GC-susceptible E. coli 136 (46.7) 1,889 (56.8)
3GC-resistant E. coli 38 (13.05) 253 (7.6)
3GC-susceptible Klebsiella spp. 38 (13.05) 310 (9.3)
3GC-resistant Klebsiella spp. 21 (7.2) 62 (1.9)
*Values are no. (%) except as indicated. p values calculated by using likelihood ratio tests. 3GC, third-generation cephalosporin; ICU, intensive care unit;
MRSA, methicillin-resistant S. aureus; MSSA, methicillin-susceptible S. aureus.
†Missing data: 9 stays with recurrence, 174 stays without recurrence.

978 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 30, No. 5, May 2024
 Antimicrobial Resistance after Bacteremia

Table 2. Univariable and multivariable analyses of risk factors for bacteremia recurrence at 1 year after an incident stay with
community-onset bacteremia attributable to Staphylococcus aureus, Escherichia coli, or Klebsiella spp., Assistance Publique–Hôpitaux
de Paris university hospital group, Paris, France, 2017–2019*
Univariable analyses Multivariable analyses
Characteristic HR (95% CI) p value HR (95% CI) p value
Patients
Sex; referent male 0.71 (0.57–0.90) 0.004 0.94 (0.73–1.20) 0.59
Age group, y; referent 35–50 y 0.0005 0.075
>18–35 0.51 (0.25–1.03) 0.66 (0.31–1.38)
50–65 1.23 (0.82–1.83) 1.15 (0.76–1.74)
65–80 1.14 (0.78–1.69) 1.13 (0.76–1.69)
>80 0.68 (0.44–1.05) 0.77 (0.49–1.20)
Charlson comorbidity index; referent 0 <0.0001
1–2 1.95 (1.46–2.62)
>2 2.31 (1.71–3.12)
Underlying conditions
Cancer 2.36 (1.86–3.00) <0.0001 2.03 (1.58–2.62) <0.0001
Heart failure 0.91 (0.64–1.30) 0.61
Diabetes 0.96 (0.73–1.27) 0.79
Vascular disease 0.75 (0.49–1.14) 0.18
Renal disease 1.60 (1.20–2.13) 0.002 1.72 (1.28–2.31) 0.0007
Liver disease 1.89 (1.35–2.65) 0.0007 1.66 (1.17–2.35) 0.007
Chronic pulmonary disease 0.71 (0.40–1.26) 0.24
Dementia 0.56 (0.29–1.09) 0.06
Paralysis (hemiplegia / paraplegia) 0.53 (0.20–1.43) 0.21
Systemic disease 1.70 (0.70–4.12) 0.30
Incident stays
Length of stay with bacteremia; referent 7–14 d 0.30
≤7 0.78 (0.59–1.03)
14–30 0.79 (0.55–1.13)
>30 0.97 (0.63–1.50)
Surgery 1.01 (0.71–1.42) 0.97
ICU admission 1.15 (0.89–1.51) 0.30
Septic shock 1.09 (0.73–1.62) 0.68
Infection source; referent urinary tract <0.0001 0.0002
None identified 2.25 (1.62–3.13) 2.26 (1.60–3.19)
Multiple sites 1.19 (0.85–1.68) 1.22 (0.85–1.75)
Lower respiratory tract 1.12 (0.63–1.98) 1.26 (0.70–2.26)
Digestive tract 1.69 (1.13–2.54) 1.57 (1.03–2.38)
Device-related 2.32 (1.44–3.74) 1.93 (1.16–3.23)
Other 0.80 (0.43–1.51) 0.98 (0.51–1.75)
Bacteria resistance; referent MSSA <0.0001 <0.0001
MRSA 0.74 (0.27–2.04) 0.79 (0.29–2.19)
3GC-susceptible E. coli 0.99 (0.72–1.36) 1.16 (0.81–1.66)
3GC-resistant E. coli 1.99 (1.31–3.01) 2.35 (1.50–3.68)
3GC-susceptible Klebsiella spp. 1.64 (1.09–2.49) 1.41 (0.91–2.21)
3GC-resistant Klebsiella spp. 4.11 (2.48–6.81) 3.91 (2.32–6.59)
*p values calculated by using Gray’s test of the subdistribution function for univariable analyses, and Fine–Gray regression models for multivariable
analyses. 3GC, third-generation cephalosporin; HR, subdistribution hazard ratio; ICU, intensive care unit; MRSA, methicillin-resistant S. aureus; MSSA,
methicillin-susceptible S. aureus.

(referent 3GC-susceptible Klebsiella spp.). For 3GC- Discussion

resistant E. coli, the HR was slightly higher for recur- In this cohort study, we have shown that 3GC resistance
rence to the same species (2.47 [95% CI 1.54–3.95]), in Klebsiella spp. or E. coli in community-onset bactere-
and slightly lower for recurrence to a different spe- mia significantly increases the probability of all-cause
cies (1.68 [95% CI 0.94–3.01]; referent 3GC-resistant bacteremia recurrence within 1 year, whereas identifi-
E. coli). Stratified analysis by bacteria found com- cation of MRSA does not affect risk for recurrence. Our
parable results, except for the HR of the association results confirm, in community-onset bacteremia, that
between 3GC-resistant Klebsiella spp. and recurrence certain patient underlying conditions (cancer, liver dis-
to the same species, which was slightly lower (2.32 ease, and renal disease) and infection sources (digestive
[95% CI 0.96–5.62]), likely attributable to reduced tract, device-related, and no identified infection source)
sample size (Appendix Table 8). Multivariate logis- are important risk factors for bacteremia recurrence. Of
tic regression models for recurrence to the same and all identified risk factors, the isolation of 3GC-resistant
to different species yielded similar associations (Ap- Klebsiella spp. was associated with the greatest increase
pendix Table 9). in the probability of recurrence over time.

Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 30, No. 5, May 2024 979
RESEARCH

Table 3. Subdistribution HRs for relationship between each bacteria-resistance pair and recurrence of bacteremia at 1 year in final
multivariable model, by reference, in study of community-onset bacteremia attributable to Staphylococcus aureus, Escherichia coli, or
Klebsiella spp., Assistance Publique–Hôpitaux de Paris university hospital group, Paris, France, 2017–2019*
HR (95% CI)
Referent 3GC-susceptible
Bacteria resistance Referent MSSA Referent 3GC-susceptible E. coli Klebsiella spp.
MSSA Referent 0.86 (0.60–1.23) 0.71 (0.45–1.11)
MRSA 0.79 (0.29–2.19) 0.68 (0.25–1.86) 0.56 (0.20–1.59)
3GC-susceptible E. coli 1.16 (0.81–1.66) Referent 0.82 (0.56–1.20)
3GC-resistant E. coli 2.35 (1.50–3.68) 2.02 (1.41–2.91) 1.66 (1.04–2.66)
3GC-susceptible Klebsiella spp. 1.41 (0.91–2.21) 1.22 (0.83–1.78) Referent
3GC-resistant Klebsiella spp. 3.91 (2.32–6.59) 3.37 (2.10–5.41) 2.77 (1.60–4.79)
*Results are adjusted on all the variables described in Table 2. 3GC, third-generation cephalosporin; HR, subdistribution hazard ratio; MRSA, methicillin-
resistant S. aureus; MSSA, methicillin-susceptible S. aureus.

Few studies have examined the relationship In agreement with Choi et al. (9), we found no effect
between AMR and bacteremia recurrence (8–13). of MRSA on recurrence after adjustment, although
Woudt et al. (8) showed an association between the our findings are not directly comparable given the
isolation of MRSA or 3GC-resistant Enterobacteria- community-onset nature of the incident stays and
ceae and recurrence of bacteremia attributable to the shorter duration of follow-up.
same species, with crude relative risks of <2. Choi et We studied recurrence up to 1 year, accounting
al. (9) found no effect of MRSA on recurrence of S. for all species and types of bacteremia onset. Al-
aureus bacteremia over a 7-year study period, after though this definition differs from previous works,
adjustment. One study has focused on the commu- which focused on recurrence to the same bacteri-
nity context and showed crude associations between um, we argue that it better captures the potential
E. coli sequence types 131 or 405, which could be used effect of AMR and ensuing antibiotic exposure on
as a proxy for AMR, and the risk for recurrence (13). host microbiota and overall susceptibility to in-
Our study has shown an effect of 3GC-resistance in fection (17). Irrespective of their appropriateness
community-onset bacteremia attributable to Kleb- to treat a given bacterium, antibiotics can induce
siella spp. or E. coli on the probability of recurrence dysbiosis, with repercussions for host immunity,
over time, after adjustment for diverse risk factors selection of antibiotic-resistant strains, coloniza-
and while considering death as a competing event. tion, and infection by antibiotic-susceptible or
antibiotic-resistant strains (17,26–28). Given that
approximately half the recurrences were attribut-
able to the same species, we conducted sensitivity
analyses in the 2 subgroups with recurrence attrib-
utable to the same or to a different species. Those
analyses showed results consistent with the overall
analysis, while suggesting a greater effect of isolat-
ing 3GC-resistant E.coli during the incident episode
on the risk for recurrence attributable to the same
species compared with recurrence attributable to a
different species. On the other hand, the link be-
tween 3GC-resistant Klebsiella spp. and the risk for
recurrence attributable to the same or a different
species was similar. Those results support previous
works showing higher rates of illness associated
with Klebsiella spp. infections compared with E. coli
infections. Al-Hasan et al. (29) showed that isola-
tion of Klebsiella spp. was associated with bactere-
Figure 2. Cumulative incidence function curves showing mia recurrence, relative to isolation of E. coli and
probability of recurrence over time for each bacteria-resistance
pair after community-onset bacteremia attributable to
after adjustment. Other work has suggested that
Staphylococcus aureus, Escherichia coli, or Klebsiella spp., patients with ESBL-producing Klebsiella pneumoniae
Assistance Publique–Hôpitaux de Paris university hospital group, bacteremia have higher rates of ICU admission and
Paris, France, 2017–2019. 3GC, third-generation cephalosporin; death compared with patients with ESBL-produc-
ICU, intensive care unit; MRSA, methicillin-resistant S. aureus; ing E. coli bacteremia (30,31). Overall, our findings
MSSA, methicillin-susceptible S. aureus.

980 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 30, No. 5, May 2024
 Antimicrobial Resistance after Bacteremia

Table 4. Subdistribution HRs for the relationship between bacteria resistance and recurrence of bacteremia at 1 year in analysis
stratified by species in study of community-onset bacteremia attributable to Staphylococcus aureus, Escherichia coli, or Klebsiella
spp., Assistance Publique–Hôpitaux de Paris university hospital group, Paris, France, 2017–2019*
HR (95% CI)
Bacteria resistance S. aureus E. coli Klebsiella spp.
Susceptible Referent Referent Referent
Resistant 0.82 (0.29–2.31) 2.08 (1.44–3.00) 2.41 (1.35–4.30)
*Results were adjusted on all variables included in the multivariable model. Susceptible category included methicillin-susceptible S. aureus, and third-
generation cephalosporin-susceptible E. coli and Klebsiella spp. Resistant category included methicillin-resistant S. aureus, and third-generation
cephalosporin-resistant E. coli and Klebsiella spp. HR, subdistribution hazard ratio.

demonstrate a new facet to the disease burden im- of community-onset incident stays (14–16) or by
posed by 3GC resistance in E. coli and especially Kleb- the fact that recurrent stays were only identified
siella spp. infections, the proportions of which are among AP-HP hospitals.
increasing in community settings worldwide (32). Despite the size of our cohort, this study is not
Our findings support ongoing calls for increased population-based, given that it covers approximate-
awareness and intervention to limit the spread of ly one quarter of all acute care inpatients in Île-de-
antibiotic-resistant E. coli and Klebsiella spp. in the France. Moreover, it included patients hospitalized
community. For clinicians in particular, our results in university hospitals, who may have more under-
highlight the need for increased caution in the lying conditions and exposures to care, affecting the
follow-up of patients with community-onset bac- risk for recurrence. To minimize this bias, we adjust-
teremia attributable to 3GC-resistant Klebsiella spp. ed our results on most previously identified risk fac-
or E. coli. tors of recurrence, which could be related to patients
In this study, we have also shown that specific or their hospital stay and infection characteristics
underlying conditions, namely cancer, liver disease, (14–16). Because data on exposure to care was only
and renal disease, are associated with recurrence, and available among AP-HP hospitals, we could not
should thus warrant special attention during patient study this risk factor and used a commonly accepted
follow-up. To date, such findings were absent for definition of community-onset infections as occur-
community-onset bacteremia, and available stud- ring within the first 48 hours of admission (15,16).
ies considering certain underlying conditions have Moreover, information on empirical treatment,
shown heterogeneous results (9–12,33,34). Further- which could affect recurrence risk, was only avail-
more, as previously observed, we underlined that able for one third of included cases because of the
the absence of an infection source or the presence of multiplicity of drug prescription software platforms
a digestive or device-related infection source were as- used across included hospitals (14,16). Despite this
sociated with recurrence (16). limitation, a sensitivity analysis on patients with in-
Strengths of our study include the large size of formation on their empirical treatment showed simi-
our cohort, as well as the richness of the clinical and lar results to the main analysis, thereby supporting
microbiologic data available, which allowed for the our findings.
evaluation of potential effects of diverse risk fac- In conclusion, we have shown that resistance
tors. Although the numbers of antibiotic-resistant to 3GCs in Klebsiella spp. and E. coli during incident
bacterial isolates in the recurrence group varied, we community-onset bacteremia significantly increases
found statistically significant results for E. coli and bacteremia recurrence risk over time. This risk was
Klebsiella spp. and were able to describe a bacteria- highest for 3GC-resistant Klebsiella spp., for which
specific effect of resistance on recurrence. More- increasing community dissemination represents an
over, it is notable that subdistribution HRs calcu- urgent public health problem. These findings reveal
lated with the Fine–Gray regression model were an important facet to the disease and death induced
very close to adjusted odds ratios calculated with by antimicrobial-resistant Enterobacteriaceae and in-
the multivariable logistic regression model, which form a need for careful follow-up of patients recover-
supports our results. A close relationship between ing from bacteremia caused by these bacteria, as well
HRs and ORs values is expected if the event studied as and a need for interventions to limit their further
has a low probability of occurrence over time (23), spread in the community.
which is the case in our study, given that the 1-year
recurrence rate of bacteremia was 8.0%. This rate Acknowledgments
was lower than that reported by other studies (9%– The authors thank the Health Data Hub staff, which
12%), which could be explained by the selection provided essential administrative and technical support

Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 30, No. 5, May 2024 981
RESEARCH

to the study, without which this database could not have References
been built. Our thanks also go to the AP-HP clinical data 1. Murray CJ, Ikuta KS, Sharara F, Swetschinski L,
Robles Aguilar G, Gray A, et al.; Antimicrobial Resistance
warehouse staff (particularly the Innovation and Data
Collaborators. Global burden of bacterial antimicrobial
team) who implemented the clinical data warehouse and resistance in 2019: a systematic analysis. Lancet.
greatly assisted us in selecting our population, building, 2022;399:629–55. PubMed https://doi.org/10.1016/
and qualifying this database. We would like to also thank S0140-6736(21)02724-0
2. Cassini A, Högberg LD, Plachouras D, Quattrocchi A,
other AP-HP staff (particularly the Direction of Clinical
Hoxha A, Simonsen GS, et al.; Burden of AMR Collaborative
Research, Innovation, and Relations with Universities and Group. Attributable deaths and disability-adjusted life-years
Research organizations, the Clinical Research Unit of Paris caused by infections with antibiotic-resistant bacteria in the
Saclay Ouest, and all hospital departments that provided EU and the European Economic Area in 2015: a population-
level modelling analysis. Lancet Infect Dis. 2019;19:56–66.
the data), and L’Institut National de la Santé et de la
https://doi.org/10.1016/S1473-3099(18)30605-4
Recherche Médicale staff (particularly the Institut National 3. Founou RC, Founou LL, Essack SY. Clinical and
de la Santé et de la Recherche Transfert, the Department of economic impact of antibiotic resistance in developing
Legal Affairs, and the Délégation Régionale Paris–Île-de- countries: a systematic review and meta-analysis. PLoS One.
2017;12:e0189621.
France Sud).
4. Abbara S, Guillemot D, El Oualydy S, Kos M, Poret C,
Data sharing statement: All pseudonymized data collected Breant S, et al. Antimicrobial resistance and mortality in
hospitalized patients with bacteremia in the Greater Paris
for the study can be made available on request from the area from 2016 to 2019. Clin Epidemiol. 2022;14:1547–60.
AP-HP clinical data warehouse staff, on the condition that https://doi.org/10.2147/CLEP.S385555
the research project is accepted by the scientific and ethics 5. Cosgrove SE. The relationship between antimicrobial
committee of the AP-HP clinical data warehouse. resistance and patient outcomes: mortality, length of hospital
stay, and health care costs. Clin Infect Dis. 2006;42(Suppl
S.A. received a doctoral fellowship from the Ecole des 2):S82–9. https://doi.org/10.1086/499406
6. de Kraker MEA, Wolkewitz M, Davey PG, Koller W,
Hautes Etudes en Santé Publique (https://www.ehesp.fr).
Berger J, Nagler J, et al. Burden of antimicrobial resistance in
Funding received by S.A. had no role in the design of the European hospitals: excess mortality and length of hospital
study, in the collection, analyses, or interpretation of data, stay associated with bloodstream infections due to
in the writing of the manuscript, or in the decision to Escherichia coli resistant to third-generation cephalosporins.
J Antimicrob Chemother. 2011;66:398–407. https://doi.org/
publish the results. L.W. received consulting fees from
10.1093/jac/dkq412
Pfizer, HEVA, and Sanofi for unrelated projects. Other 7. Stewardson AJ, Allignol A, Beyersmann J, Graves N,
authors report no competing interests. Schumacher M, Meyer R, et al.; TIMBER Study Group. The
health and economic burden of bloodstream infections
Author contributions: S.A., D.G., C.B., and L.W. caused by antimicrobial-susceptible and non-susceptible
conceptualized and determined the methodology for the Enterobacteriaceae and Staphylococcus aureus in European
study. L.W. was the scientific manager and main hospitals, 2010 and 2011: a multicentre retrospective cohort
study. Euro Surveill. 2016;21:30319. https://doi.org/
supervisor. S.B. ran the requests to identify patients with 10.2807/1560-7917.ES.2016.21.33.30319
bacteremia within the AP-HP clinical data warehouse 8. Woudt SHS, de Greeff SC, Schoffelen AF, Vlek ALM,
and worked with M.K., S.A., and S.E.O. to qualify the Bonten MJM, Cohen Stuart JWT, et al.; Infectious Diseases
data. M.K., S.A., and S.E.O. verified, qualified, cleaned, Surveillance Information System–Antimicrobial Resistance
(ISIS-AR) Study Group. Antibiotic resistance and the risk
and structured the database. During the construction of of recurrent bacteremia. Clin Infect Dis. 2018;66:1651–7.
the database, C.P. supervised the project on the AP-HP https://doi.org/10.1093/cid/cix1076
clinical data warehouse side. S.A. selected the population 9. Choi SH, Dagher M, Ruffin F, Park LP, Sharma-Kuinkel BK,
for this study within the database, did the analyses, and Souli M, et al. Risk factors for recurrent Staphylococcus aureus
bacteremia. Clin Infect Dis. 2021;72:1891–9. https://doi.org/
wrote the first draft of the paper. S.A., C.B., D.R.M.S., 10.1093/cid/ciaa801
and L.W. edited and revised the manuscript until the 10. Turett GS, Blum S, Telzak EE. Recurrent pneumococcal
final version. S.A., L.W., S.E.O., M.K., and S.B. had access bacteremia: risk factors and outcomes. Arch Intern Med. 2001;
to the study data. All authors have seen and approved 161:2141–4. https://doi.org/10.1001/archinte.161.17.2141
11. Lee CH, Su LH, Chen FJ, Tang YF, Chien CC, Liu JW.
the final manuscript. Clinical and microbiologic characteristics of adult patients
with recurrent bacteraemia caused by extended-spectrum
β-lactamase-producing Escherichia coli or Klebsiella
About the Author pneumoniae. Clin Microbiol Infect. 2015;21:1105.e1–8.
Dr. Abbara is an infectious diseases physician and https://doi.org/10.1016/j.cmi.2015.07.025
12. Harris PNA, Peri AM, Pelecanos AM, Hughes CM,
epidemiologist. Her primary research interests are Paterson DL, Ferguson JK. Risk factors for relapse or
antimicrobial resistance and community-onset infections, persistence of bacteraemia caused by Enterobacter spp.:
with a focus on the contribution of hospital clinical a case-control study. Antimicrob Resist Infect Control.
data warehouses. 2017;6:14. https://doi.org/10.1186/s13756-017-0177-0

982 Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 30, No. 5, May 2024
 Antimicrobial Resistance after Bacteremia

13. Fröding I, Hasan B, Sylvin I, Coorens M, Nauclér P, 24. Therneau TM. A package for survival analysis in R.
Giske CG. Extended-spectrum-β-lactamase- and plasmid 2020 [cited 2024 Apr 3]. https://CRAN.R-project.org/
AmpC-producing Escherichia coli causing community-onset package=survival
bloodstream infection: association of bacterial clones and 25. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC,
virulence genes with septic shock, source of infection, and Vandenbroucke JP, et al. The strengthening the reporting
recurrence. Antimicrob Agents Chemother. 2020;64:e02351– of observational studies in epidemiology (STROBE)
19. https://doi.org/10.1128/AAC.02351-19 statement: guidelines for reporting observational studies.
14. Gradel KO, Jensen US, Schønheyder HC, Østergaard C, Lancet. 2007;370:1453–7.
Knudsen JD, Wehberg S, et al.; Danish Collaborative 26. Dethlefsen L, Relman DA. Incomplete recovery and
Bacteraemia Network (DACOBAN). Impact of appropriate individualized responses of the human distal gut microbiota
empirical antibiotic treatment on recurrence and mortality to repeated antibiotic perturbation. Proc Natl Acad Sci U
in patients with bacteraemia: a population-based cohort S A. 2011;108(Suppl 1):4554–61. https://doi.org/10.1073/
study. BMC Infect Dis. 2017;17:122. https://doi.org/10.1186/ pnas.1000087107
s12879-017-2233-z 27. Zhang M, Jiang Z, Li D, Jiang D, Wu Y, Ren H, et al. Oral
15. Jensen US, Knudsen JD, Wehberg S, Gregson DB, antibiotic treatment induces skin microbiota dysbiosis and
Laupland KB. Risk factors for recurrence and death after influences wound healing. Microb Ecol. 2015;69:415–21.
bacteraemia: a population-based study. Clin Microbiol https://doi.org/10.1007/s00248-014-0504-4
Infect. 2011;17:1148–54. https://doi.org/10.1111/ 28. Bhalodi AA, van Engelen TSR, Virk HS, Wiersinga WJ.
j.1469-0691.2011.03587.x Impact of antimicrobial therapy on the gut microbiome.
16. Jensen US, Knudsen JD, Østergaard C, Gradel KO, J Antimicrob Chemother. 2019;74(Suppl 1):i6–15.
Frimodt-Møller N, Schønheyder HC. Recurrent bacteraemia: https://doi.org/10.1093/jac/dky530
a 10-year regional population-based study of clinical and 29. Al-Hasan MN, Eckel-Passow JE, Baddour LM. Recurrent
microbiological risk factors. J Infect. 2010;60:191–9. gram-negative bloodstream infection: a 10-year population-
https://doi.org/10.1016/j.jinf.2009.12.007 based cohort study. J Infect. 2010;61:28–33. https://doi.org/
17. Smith DR, Temime L, Opatowski L. Microbiome-pathogen 10.1016/j.jinf.2010.03.028
interactions drive epidemiological dynamics of antibiotic 30. Scheuerman O, Schechner V, Carmeli Y, Gutiérrez-Gutiérrez B,
resistance: a modeling study applied to nosocomial pathogen Calbo E, Almirante B, et al.; REIPI/ESGBIS/INCREMENT
control. eLife. 2021;10:e68764. https://doi.org/10.7554/ investigators. Comparison of predictors and mortality
eLife.68764 between bloodstream infections caused by ESBL-producing
18. Titelman E, Hasan CM, Iversen A, Nauclér P, Kais M, Escherichia coli and ESBL-producing Klebsiella pneumoniae.
Kalin M, et al. Faecal carriage of extended-spectrum Infect Control Hosp Epidemiol. 2018;39:660–7.
β-lactamase-producing Enterobacteriaceae is common 12 https://doi.org/10.1017/ice.2018.63
months after infection and is related to strain factors. 31. Burnham JP, Kwon JH, Olsen MA, Babcock HM, Kollef MH.
Clin Microbiol Infect. 2014;20:O508–15. https://doi.org/ Differences in mortality between infections due to extended-
10.1111/1469-0691.12559 spectrum-beta-lactamase-producing Klebsiella pneumoniae and
19. Bezabih YM, Sabiiti W, Alamneh E, Bezabih A, Peterson GM, Escherichia coli. Infect Control Hosp Epidemiol. 2018;39:1138–
Bezabhe WM, et al. The global prevalence and trend of 9. https://doi.org/10.1017/ice.2018.142
human intestinal carriage of ESBL-producing Escherichia coli 32. Chang D, Sharma L, Dela Cruz CS, Zhang D. Clinical
in the community. J Antimicrob Chemother. 2021;76:22–9. epidemiology, risk factors, and control strategies of Klebsiella
https://doi.org/10.1093/jac/dkaa399 pneumoniae infection. Front Microbiol. 2021;12:750662.
20. Coque TM, Baquero F, Cantón R. Increasing prevalence of https://doi.org/10.3389/fmicb.2021.750662
ESBL-producing Enterobacteriaceae in Europe. Euro Surveill. 33. Albertson J, McDanel JS, Carnahan R, Chrischilles E,
2008;13:19044. https://doi.org/10.2807/ese.13.47.19044-en Perencevich EN, Goto M, et al. Determination of risk factors
21. Gauzit R, Castan B, Bonnet E, Bru JP, Cohen R, Diamantis S, for recurrent methicillin-resistant Staphylococcus aureus
et al. Anti-infectious treatment duration: the SPILF and GPIP bacteremia in a Veterans Affairs healthcare system
French guidelines and recommendations. Infect Dis Now. population. Infect Control Hosp Epidemiol. 2015;36:543–9.
2021;51:114–39. https://doi.org/10.1016/j.idnow.2020.12.001 https://doi.org/10.1017/ice.2015.25
22. Comité de l’Antibiogramme de la Société Francaise de 34. Sanz-García M, Fernández-Cruz A, Rodríguez-Créixems M,
Microbiologie–European Committee on Antimicrobial Cercenado E, Marin M, Muñoz P, et al. Recurrent
Susceptibility Testing. Guidelines of the Comité de Escherichia coli bloodstream infections: epidemiology and risk
l’Antibiogramme de la Société Francaise de Microbiologie– factors. Medicine (Baltimore). 2009;88:77–82. https://doi.org/
European Committee on Antimicrobial Susceptibility Testing 10.1097/MD.0b013e31819dd0cf
[cited 2024 Apr 3]. https://www.sfm-microbiologie.org/
casfm Address for correspondence: Salam Abbara, Institut Pasteur
23. Austin PC, Fine JP. Practical recommendations for reporting
Paris, 25 rue du Docteur Roux, 75015 Paris, France; email:
Fine–Gray model analyses for competing risk data. Stat Med.
2017;36:4391–400. https://doi.org/10.1002/sim.7501 [email protected]

Emerging Infectious Diseases • www.cdc.gov/eid • Vol. 30, No. 5, May 2024 983
Article DOI: https://doi.org/10.3201/eid3005.231555

Antimicrobial Resistance as Risk Factor for

Recurrent Bacteremia after Staphylococcus
aureus, Escherichia coli, or Klebsiella spp.
Community-Onset Bacteremia
Appendix
Appendix Table 1. Most frequently isolated bacteria in first recurrent episodes, stratified by the bacterial species isolated in incident
episodes.
Incident episode due to Incident episode due to Incident episode due
E. coli Klebsiella spp. to S. aureus
Bacteria, first recurrent episodes, no. (no. = 174) (no. = 59) (no. = 58)
E. coli 92 9 8
Polymicrobial 25 8 6
Klebsiella spp. 14 25 3
S. aureus 6 3 21
P. aeruginosa 5 5 4
Others 32 9 16

Appendix Table 2. Concordance between the bacterial resistance phenotype of incident and first recurrent episodes, among
recurrences due to the same species as the incident episode.
Bacteria-resistance Same phenotype, no./NO.
E. coli
3GC-S 63/68
3GC-R 23/24
Klebsiella spp.
3GC-S 14/15
3GC-R 8/10
S. aureus
Methicillin-S 19/20
Methicillin-R 1/1
3GC-S / -R: susceptible / resistant to 3rd-generation cephalosporins; methicillin-S / -R: methicillin- susceptible / -resistant.

Appendix Table 3. Multivariable logistic regression models of risk factors for bacteremia recurrence at 1 y, following an incident
stay with community-onset bacteremia due to S. aureus, E. coli, or Klebsiella spp., 2017–2019
Recurrence to any species
Characteristic aOR [95CI] P-value
Patients
Sex (ref = male) 0.93 [0.71 – 1.21] 0.56
Age, years (ref = [35–50]) 0.06
[18–35] 0.62 [0.27 – 1.30]
[50–65] 1.14 [0.74 – 1.80]
[65–80] 1.12 [0.73 – 1.75]
>80 0.76 [0.47 – 1.23]
Comorbidities
Cancer (ref = absence) 2.19 [1.66 – 2.88] <0.0001
Renal disease (ref = absence) 1.78 [1.28 – 2.45] 0.0005
Liver disease (ref = absence) 1.80 [1.21 – 2.62] 0.006
Incident stays
Infection source (ref = urinary tract) 0.0003
None identified 2.41 [1.66 – 3.49]
Multiple sites 1.24 [0.84 – 1.81]

Page 1 of 5
 Recurrence to any species
Characteristic aOR [95CI] P-value
Lower respiratory tract 1.25 [0.65 – 2.24]
Digestive tract 1.61 [1.02 – 2.49]
Device-related 1.97 [1.10 – 3.41]
Other 0.98 [0.47 – 1.89]
Bacteria-resistance (ref = MSSA) <0.0001
MRSA 0.77 [0.23 – 1.99]
3GC-S E. coli 1.15 [0.79 – 1.68]
3GC-R E. coli 2.48 [1.53 – 4.01]
3GC-S Klebsiella spp. 1.43 [0.88 – 2.28]
3GC-R Klebsiella spp. 4.70 [2.52 – 8.58]
P-values were calculated using multivariable logistic regression models. Abbreviations: aOR: adjusted odds ratio; 3GC-S / 3GC-R: 3rd-generation
cephalosporin-susceptible / -resistant; ICU: intensive care unit; MSSA / MRSA: methicillin-susceptible / -resistant S. aureus.

Appendix Table 4. Descriptive and Fine-Gray multivariable analyses of risk factors for bacteremia recurrence at 1 y, following an
incident stay with community-onset bacteremia due to S. aureus, E. coli, or Klebsiella spp. in the subsample of incident stays
including information on the adequacy of empirical treatment, 2017–2019.
Description of the population, N (%) Multivariable analyses
With recurrence Without recurrence
Characteristic N = 117 N = 1175 HR [95CI] P-value
Patients
Sex, no. (%) 0.45
Male 67 (57.3) 522 (44.4) 1
Female 50 (42.7) 653 (55.6) 0.86 [0.58 - 1.27]
Age, years, no. (%) 0.006
[18–35] 1 (0.9) 80 (6.8) 0.17 [0.02 - 1.32]
[35–50] 12 (10.3) 130 (11.1) 1
[50–65] 39 (33.3) 254 (21.6) 1.58 [0.82 - 3.05]
[65–80] 44 (37.6) 392 (33.4) 1.10 [0.57 - 2.12]
>80 21 (17.9) 319 (27.1) 0.75 [0.36 - 1.55]
Comorbidities*, no. (%)
Cancer 40 (34.5) 218 (19) 1.73 [1.14 - 2.62] 0.01
Renal disease 25 (21.6) 171 (14.9) 1.63 [1.02 - 2.59] 0.05
Liver disease 16 (13.8) 82 (7.2) 1.62 [0.93 - 2.82] 0.11
Incident stays
Infection source*, no. (%) 0.005
None identified 25 (21.6) 127 (11.1) 3.04 [1.75 - 5.29]
Multiple sites 27 (23.3) 262 (22.9) 1.72 [0.99 - 2.97]
Lower respiratory tract 6 (5.2) 47 (4.1) 2.38 [0.96 - 5.92]
Urinary tract 30 (25.9) 481 (42) 1
Digestive tract 11 (9.5) 104 (9.1) 1.60 [0.79 - 3.25]
Device-related 12 (10.3) 52 (4.5) 3.20 [1.54 - 6.69]
Other 5 (4.3) 73 (6.4) 1.75 [0.63 - 4.84]
Bacteria-resistance, no. (%) 0.003
MSSA 17 (14.5) 205 (17.4) 1
MRSA 2 (1.7) 24 (2) 0.93 [0.21 - 4.09]
3GC-S E. coli 54 (46.2) 705 (60) 1.44 [0.79 - 2.63]
3GC-R E. coli 20 (17.1) 113 (9.6) 2.84 [1.33 - 6.09]
3GC-S Klebsiella spp. 13 (11.1) 105 (8.9) 1.41 [0.65 - 3.05]
3GC-R Klebsiella spp. 11 (9.4) 23 (2) 6.57 [2.69 - 16.06]
Adequacy of empiric treatment, no. (%) 0.51
Appropriate 95 (81.2) 1072 (91.2) 1
Inappropriate 22 (18.8) 103 (8.8) 1.22 [0.68 - 2.17]
Abbreviations: 3GC-S / 3GC-R: susceptible / resistant to 3rd-generation cephalosporins; HR: subdistribution hazard ratios; ICU: intensive care unit;
MSSA / MRSA: methicillin-susceptible / resistant S. aureus. *Missing data: 1 stay with recurrence, 29 stays without recurrence.

Page 2 of 5
Appendix Table 5. Univariable and multivariable analyses of risk factors for bacteremia recurrence to the same species at 1 y,
following an incident stay with community-onset bacteremia due to S. aureus, E. coli, or Klebsiella spp., 2017–2019.
Univariable analyses Multivariable analyses
Characteristic HR [95CI] P-value HR [95CI] P-value
Patients
Sex (ref = male) 0.75 [0.54 – 1.06] 0.1 0.90 [0.63 – 1.29] 0.56
Age (ref = [35–50]) 0.07 0.14
≥18–35] 0.45 [0.15 – 1.36] 0.44 [0.13 – 1.53]
[50–65] 1.29 [0.71 – 1.33] 1.30 [0.71 – 2.37]
[65–80] 1.23 [0.69 – 1.18] 1.19 [0.66 – 2.13]
>80 0.81 [0.43 – 1.51] 0.82 [0.43 – 1.57]
Charlson comorbidity index (ref = 0) 0.02
1–2 1.65 [1.11 – 2.46]
>2 1.60 [1.04 – 2.47]
Comorbidities
Cancer 1.52 [1.04 – 2.22] 0.03
Heart failure 0.92 [0.56 – 1.53] 0.76
Diabetes 1.03 [0.69 – 1.53] 0.90
Vascular disease 0.88 [0.50 – 1.56] 0.66
Renal disease 2.16 [1.47 – 3.17] <0.0001 2.04 [1.37 – 3.03] 0.0009
Liver disease 1.94 [1.19 – 3.14] 0.008 1.70 [1.03 – 2.80] 0.05
Chronic pulmonary disease 0.36 [0.12 – 1.13] 0.08
Dementia 0.51 [0.19 – 1.39] 0.19
Paralysis (hemiplegia / paraplegia) 0.55 [0.14 – 2.21] 0.40
Systemic disease 0.72 [0.10 – 5.17] 0.75
Incident stays
Length of stay with bacteremia (ref = [7–14]) 0.30
≤7 0.74 [0.50 – 1.10]
[14–30] 0.64 [0.38 – 1.10]
>30 0.74 [0.38 – 1.45]
Surgery 1.15 [0.71 – 1.84] 0.58
ICU admission 1.17 [0.80 – 1.73] 0.42
Septic shock 1.09 [0.62 – 1.93] 0.77
Infection source (ref = urinary tract) 0.20 0.17
None identified 1.49 [0.92 – 2.39] 1.87 [1.14 – 3.06]
Multiple sites 1.88 [0.55 – 1.42] 1.06 [0.64 – 1.74]
Lower respiratory tract 0.65 [0.26 – 1.64] 0.84 [0.33 – 2.15]
Digestive tract 1.37 [1.78 – 2.38] 1.48 [0.83 – 2.62]
Device-related 1.41 [1.67 – 2.99] 1.90 [0.86 – 4.20]
Other 0.70 [0.30 – 1.63] 1.10 [0.45 – 2.71]
Bacteria-resistance (ref = MSSA) <0.0001 <0.0001
MRSA 0.49 [0.07 – 3.68] 0.53 [0.07 – 3.95]
3GC-S E. coli 1.33 [0.81 – 2.19] 1.68 [0.95 – 2.95]
3GC-R E. coli 3.41 [1.89 – 6.18] 4.13 [2.14 – 7.98]
3GC-S Klebsiella spp 1.77 [0.90 – 3.45] 1.85 [0.91 – 3.73]
3GC-R Klebsiella spp 5.55 [2.60 – 11.86] 5.62 [2.54 – 12.41]
P-values were calculated using Gray’s test of the subdistribution function for univariable analyses, and Fine-Gray regression models for multivariable
analyses. Abbreviations: 3GC-S / 3GC-R: 3rd-generation cephalosporin-susceptible / -resistant; HR: subdistribution hazard ratios; ICU: intensive care
unit; MSSA / MRSA: methicillin-susceptible / -resistant S. aureus.

Appendix Table 6. Univariable and multivariable analyses of risk factors for bacteremia recurrence to a different species at 1 y,
following an incident stay with community-onset bacteremia due to S. aureus, E. coli, or Klebsiella spp., 2017–2019.
Univariable analyses Multivariable analyses
Characteristic HR [95CI] P-value HR [95CI] P-value
Patients
Sex (ref = male) 0.67 [0.48 – 0.92] 0.01 0.92 [0.65 – 1.30] 0.63
Age (ref = [35–50]) 0.01 0.46
≥18–35] 0.54 [0.21 – 1.34] 0.90 [0.35 – 2.29]
[50–65] 1.17 [0.69 – 2.01] 1.09 [0.61 – 1.92]
[65–80] 1.08 [0.64 – 1.81] 1.16 [0.67 – 2.03]
>80 0.57 [0.31 – 1.03] 0.75 [0.40 – 1.42]
Charlson comorbidity index (ref = 0) <0.0001
1–2 2.45 [1.58 – 3.80]
>2 3.40 [2.20 – 5.25]
Comorbidities
Cancer 3.56 [2.57 – 4.92] <0.0001 2.86 [2.03 – 4.03] <0.0001
Heart failure 0.91 [0.55 – 1.48] 0.69
Diabetes 0.90 [0.60 – 1.33] 0.59

Page 3 of 5
 Univariable analyses Multivariable analyses
Characteristic HR [95CI] P-value HR [95CI] P-value
Vascular disease 0.62 [0.32 – 1.17] 0.14
Renal disease 1.17 [0.75 – 1.83] 0.48
Liver disease 1.92 [1.20 – 3.08] 0.007 1.73 [1.06 – 2.81] 0.04
Chronic pulmonary disease 1.03 [0.52 – 2.01] 0.94
Dementia 0.60 [0.25 – 1.46] 0.26
Paralysis (hemiplegia / paraplegia) 0.52 [0.13 – 2.08] 0.35
Systemic disease 2.65 [0.98 – 7.17] 0.05 3.26 [1.18 – 8.98] 0.05
Incident stays
Length of stay with bacteremia (ref = [7–14]) 0.50
≤7 0.81 [0.54 – 1.21]
[14–30] 0.93 [0.57 – 1.52]
>30 1.20 [0.68 – 2.14]
Surgery 0.88 [0.53 – 1.45] 0.61
ICU admission 1.15 [0.79 – 1.67] 0.48
Septic shock 1.09 [0.63 – 1.89] 0.77
Infection source (ref = urinary tract) <0.0001 0.0007
None identified 3.60 [2.23 – 5.82] 3.04 [1.85 – 5.00]
Multiple sites 1.71 [1.03 – 2.84] 1.49 [0.87 – 2.55]
Lower respiratory tract 1.89 [0.89 – 3.99] 1.90 [0.88 – 4.11]
Digestive tract 2.27 [1.24 – 4.15] 1.83 [0.99 – 3.38]
Device-related 3.96 [2.08 – 7.52] 2.40 [1.19 – 4.82]
Other 0.96 [0.37 – 2.48] 0.99 [0.37 – 2.66]
Bacteria-resistance (ref = MSSA) <0.0001 0.01
MRSA 0.88 [0.27 – 2.86] 0.93 [0.28 – 3.05]
3GC-S E. coli 0.79 [0.52 – 1.19] 0.91 [0.57 – 1.44]
3GC-R E. coli 1.19 [0.64 – 2.22] 1.52 [0.79 – 2.93]
3GC-S Klebsiella spp 1.59 [0.94 – 2.71] 1.22 [0.68 – 2.17]
3GC-R Klebsiella spp 3.63 [1.84 – 7.17] 3.51 [1.75 – 7.06]
P-values were calculated using Gray’s test of the subdistribution function for univariable analyses, and Fine-Gray regression models for multivariable
analyses. Abbreviations: 3GC-S / 3GC-R: 3rd-generation cephalosporin-susceptible / -resistant; HR: subdistribution hazard ratios; ICU: intensive care
unit; MSSA / MRSA: methicillin-susceptible / -resistant S. aureus.

Appendix Table 7. Subdistribution hazard ratios and 95% confidence intervals for the relationship between each bacteria-
resistance pair and recurrence of bacteremia at 1 y in the final multivariable model as a function of the recurrence type, by
reference*, 2017–2019
HR HR HR
Bacteria-resistance [ref = MSSA] [ref = 3GC-S E. coli] [ref = 3GC-S Klebsiella spp.]
Recurrence to any species
MSSA 1 0.86 [0.60 – 1.23] 0.71 [0.45 – 1.11]
MRSA 0.79 [0.29 – 2.19] 0.68 [0.25 – 1.86] 0.56 [0.20 – 1.59]
3GC-S E. coli 1.16 [0.81 – 1.66] 1 0.82 [0.56 – 1.20]
3GC-R E. coli 2.35 [1.50 – 3.68] 2.02 [1.41 – 2.91] 1.66 [1.04 – 2.66]
3GC-S Klebsiella spp. 1.41 [0.91 – 2.21] 1.22 [0.83 – 1.78] 1
3GC-R Klebsiella spp. 3.91 [2.32 – 6.59] 3.37 [2.10 – 5.41] 2.77 [1.60 – 4.79]
Recurrence to the same
species
MSSA 1 0.60 [0.34 – 1.05] 0.54 [0.27 – 1.10]
MRSA 0.53 [0.07 – 3.95] 0.31 [0.04 – 2.31] 0.29 [0.04 – 2.19]
3GC-S E. coli 1.68 [0.95 – 2.95] 1 0.91 [0.51 – 1.61]
3GC-R E. coli 4.13 [2.14 – 7.98] 2.47 [1.54 – 3.95] 2.24 [1.15 – 4.35]
3GC-S Klebsiella spp. 1.85 [0.91 – 3.73] 1.10 [0.62 – 1.95] 1
3GC-R Klebsiella spp. 5.62 [2.54 – 12.41] 3.35 [1.69 – 6.63] 3.05 [1.35 – 6.86]
Recurrence to a different
species
MSSA 1 1.10 [0.70 – 1.75] 0.82 [0.46 – 1.47]
MRSA 0.93 [0.28 – 3.05] 1.03 [0.32 – 3.33] 0.77 [0.22 – 2.62]
3GC-S E. coli 0.91 [0.57 – 1.44] 1 0.74 [0.45 – 1.24]
3GC-R E. coli 1.52 [0.79 – 2.93] 1.68 [0.94 – 3.01] 1.25 [0.63 – 2.50]
3GC-S Klebsiella spp. 1.22 [0.68 – 2.17] 1.34 [0.81 – 2.24] 1
3GC-R Klebsiella spp. 3.51 [1.75 – 7.06] 3.88 [2.01 – 7.48] 2.88 [1.37 – 6.07]
* Results are adjusted on all the variables described in Table 2, Appendix Table 5 and 6 for recurrence to any species, recurrence to the same
species, and recurrence to a different species, respectively. Abbreviations: 3GC-S / 3GC-R: susceptible / resistant to 3rd-generation cephalosporins;
HR: subdistribution hazard ratios; MSSA / MRSA: methicillin-susceptible / -resistant S. aureus; OR: odds ratios.

Page 4 of 5
Appendix Table 8. Count of bacterial isolates, and subdistribution hazard ratios with 95% confidence intervals for the relationship
between antimicrobial resistance and recurrence of bacteremia at 1 y as a function of the recurrence type, in an analysis stratified
by species*, 2017–2019
S. aureus E. coli Klebsiella spp.
No. with / No. with / No. with /
without without without
Characteristic recurrence HRs recurrence HRs recurrence HRs
Recurrence to any species
Susceptible 54/737 1 136/1889 1 38/310 1
Resistant 4/75 0.82 [0.29 – 2.31] 38/253 2.08 [1.44 - 3.00] 21/62 2.41 [1.35 - 4.30]
Recurrence to the same
species
Susceptible 20/737 1 68/1889 1 15/310 1
Resistant 1/75 0.49 [0.06 – 3.80] 24/253 2.59 [1.61 - 4.17] 10/62 2.32 [0.96 - 5.62]
Recurrence to a different
species
Susceptible 34/737 1 68/1889 1 23/310 1
Resistant 3/75 0.99 [0.30 – 3.29] 14/253 1.67 [0.93 - 3.01] 11/62 2.93 [1.35 - 6.36]
The susceptible category included methicillin-susceptible S. aureus, and 3rd-generation cephalosporin susceptible E. coli and Klebsiella spp. The
resistant category included methicillin-resistant S. aureus, and 3rd-generation cephalosporin-resistant E. coli and Klebsiella spp. *For each bacteria,
results were adjusted on all variables included in the multivariable model of the recurrence group.

Appendix Table 9. Multivariable logistic regression models of risk factors for bacteremia recurrence at 1 y, following an incident
stay with community-onset bacteremia due to S. aureus, E. coli, or Klebsiella spp., 2017–2019, by recurrence type
Recurrence to the same species Recurrence to a different species
Characteristic aOR [95CI] P-value aOR [95CI] P-value
Patients
Sex (ref = male) 0.90 [0.62 - 1.30] 0.56 0.92 [0.64 - 1.31] 0.64
Age, years (ref = [35–50]) 0.14 0.47
[18–35] 0.42 [0.10 - 1.31] 0.86 [0.30 - 2.16]
[50–65] 1.28 [0.70 - 2.46] 1.09 [0.61 - 2.04]
[65–80] 1.18 [0.66 - 2.22] 1.16 [0.66 - 2.14]
>80 0.81 [0.42 - 1.62] 0.75 [0.39 - 1.47]
Comorbidities
Cancer - - 3.02 [2.11 - 4.32] <0.0001
Renal disease 2.08 [1.35 - 3.12] 0.0006 - -
Liver disease 1.80 [1.04 - 2.99] 0.04 1.77 [1.03 - 2.93] 0.04
Systemic disease - 3.57 [1.01 - 9.73] 0.048
Incident stays
Infection source (ref = urinary tract) 0.17 0.0006
None identified 1.91 [1.13 - 3.16] 3.24 [1.93 - 5.48]
Multiple sites 1.06 [0.63- 1.76] 1.52 [0.88 - 2.65]
Lower respiratory tract 0.83 [0.28 - 1.99] 1.93 [0.82 - 4.12]
Digestive tract 1.53 [0.83 - 2.73] 1.84 [0.95 - 3.44]
Device-related 1.86 [0.76 - 4.07] 2.48 [1.16 - 5.11]
Other 1.10 [0.40 - 2.61] 1.01 [0.33 - 2.55]
Bacteria-resistance (ref = MSSA) <0.0001 0.01
MRSA 0.53 [0.03 - 2.63] 0.92 [0.21 - 2.71]
3GC-S E. coli 1.66 [0.95 - 3.03] 0.90 [0.56 - 1.47]
3GC-R E. coli 4.24 [2.15 - 8.45] 1.55 [0.76 - 3.02]
3GC-S Klebsiella spp. 1.86 [0.89 - 3.82] 1.22 [0.65 - 2.21]
3GC-R Klebsiella spp. 6.17 [2.55 - 14.21] 3.99 [1.76 - 8.54]
P-values were calculated using multivariable logistic regression models. Abbreviations: aOR: adjusted odds ratio; 3GC-S / 3GC-R: 3rd-generation
cephalosporin-susceptible / -resistant; ICU: intensive care unit; MSSA / MRSA: methicillin-susceptible / -resistant S. aureus.

Page 5 of 5