University of Southern California Dissertations and Theses

Antibiotic resistance in a large ICU cohort: 1995-2002

(USC Thesis Other)

Antibiotic resistance in a large ICU cohort: 1995-2002

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

ANTIBIOTIC RESISTANCE IN A LARGE ICU COHORT: 1995-2002

by
Shivani Aggarwal

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(EPIDEMIOLOGY)

May 2019

Copyright 2019 Shivani Aggarwal
ii

Epigraph
“Had I the heavens’ embroidered cloths,
Enwrought with golden and silver light,
The blue and the dim and the dark cloths
Of night and light and the half light,
I would spread the cloths under your feet:
But I, being poor, have only my dreams;
I have spread my dreams under your feet;
Tread softly because you tread on my dreams.”
- W. B. Yeats
iii

Acknowledgements
I would like to thank Dr. Robert Larsen and my dissertation committee, Dr.
Victoria K. Cortessis, Dr. Roberta McKean-Cowdin, Dr. Wendy Cozen, Dr.
Thomas Mack, and Dr. Emi Minejima for their guidance and support.

iv

Table of Contents
Epigraph ............................................................................................................... ii
Acknowledgements ............................................................................................ iii
List of Tables ...................................................................................................... vi
List of Figures ................................................................................................... viii
Abbreviations ...................................................................................................... ix
Abstract ............................................................................................................... xi
Chapter 1: Background ...................................................................................... 1
1.1 Introduction ................................................................................................. 2
1.2 Hospital-acquired Infections ........................................................................ 3
1.3 Characteristics of Bacteria in the Hospital ................................................... 6
1.4 The Gut Microbiome and Human Health ..................................................... 8
1.5 Mechanisms of Antibiotic Resistance ........................................................ 12
1.6 The Antibiotic Resistome in Humans ......................................................... 15
1.7 Antibiotic Use Perturbs the Bacterial Distribution of the Gut and Promotes
Emergence of Resistance ............................................................................... 18
1.8 Antibiotic Metabolism and Elimination ....................................................... 20
1.9 Specific Aims ............................................................................................. 23
1.10 Figures .................................................................................................... 25
1.11 References .............................................................................................. 27
Chapter 2: Assembly of the Hospital-Based Cohort ..................................... 34
2.1 Abstract ..................................................................................................... 35
2.2 Introduction ............................................................................................... 35
2.3 Data Collection .......................................................................................... 36
2.4 Process Flow ............................................................................................. 37
2.5 Patients in the ICU .................................................................................... 38
2.6 Patients with Antibiotic Use ....................................................................... 41
2.7 Patients with Positive Blood Cultures ........................................................ 44
2.8 Patients in ICU with Antibiotic and Resistance Classifications .................. 47
2.8.1 Patients in ICU with Antibiotic Classifications ..................................... 48
2.8.2 Patients in the ICU with Resistance Score Classification .................... 55
2.9 Patients in ICU with Comorbidity Classifications ....................................... 62
v

2.10 Patients in the Final Analytic Dataset ...................................................... 63
2.11 Antibiotics and Resistance Patterns among Patients in the Final Analytic
Cohort ............................................................................................................. 65
2.12 Figures .................................................................................................... 67
2.13 References .............................................................................................. 70
Chapter 3: Patterns of Antibiotic Resistance in Organisms Cultured from
HA-BSIs in the ICU: a Hospital-Based Cohort Study ..................................... 71
3.1 Abstract ..................................................................................................... 72
3.2 Introduction ............................................................................................... 73
3.3 Materials and Methods .............................................................................. 74
3.4 Results ...................................................................................................... 84
3.5 Discussion ................................................................................................. 97
3.6 Tables and Figures .................................................................................. 103
3.9 References .............................................................................................. 120
3.10 Supplementary Material ........................................................................ 124
Chapter 4: Route of Antibiotic Elimination and Subsequent Antibiotic
Resistance....................................................................................................... 145
4.1 Abstract ................................................................................................... 147
4.2 Introduction ............................................................................................. 147
4.3 Materials and Methods ............................................................................ 149
4.4 Results .................................................................................................... 154
4.5 Discussion ............................................................................................... 157
4.6 Tables and Figures .................................................................................. 161
4.7 References .............................................................................................. 169
4.8 Supplementary Material .......................................................................... 173
Chapter 5: Conclusion ................................................................................... 178
5.1 Findings ................................................................................................... 179
5.2 Strengths and Limitations ........................................................................ 180
5.3 Future Directions ..................................................................................... 182
5.4 References .............................................................................................. 183

vi

List of Tables
Table 3.1. Baseline and Demographic Characteristics of LA-USC MC ICU
Patients, 1995-2002 ............................................................................... 103
Table 3.2. Hospitalization Characteristics of LA-USC MC ICU Patients, 1995-
2002 ....................................................................................................... 105
Table 3.3. Microbiological Characteristics of LA-USC MC ICU Patients, 1995-
2002 ....................................................................................................... 106
Table 3.4a. Association between Antibiotic Elimination Route and Drug
Resistance of HA-BSI Organisms in ICU Patients using Multinomial
Model, 1995-2002 .................................................................................. 107
Table 3.4b. Association between Antibiotic Elimination Route and Drug
Resistance of HA-BSI ESKAPE Organisms in ICU Patients using
Multinomial Model, 1995-2002 ............................................................... 108
Table 3.5. Association between Antibiotic Elimination Route and Drug Resistance
of HA-BSI Organisms in ICU Patients using Cumulative Model, 1995-2002
............................................................................................................... 109
Table 3.6. Association between Antibiotic Elimination Route and Drug Resistance
of HA-BSI Organisms Among ICU Patients with Antibiotic Use, 1995-2002
............................................................................................................... 110
Table 3.7. Antibiotic-Organism Associations of HA-BSI Organisms in ICU
Patients using Multinomial Model, 1995-2002 ....................................... 111
Table 3.8. Fraction of Drug Resistance Attributable to Antibiotic Use in ICU
Patients, 1995-2002 ............................................................................... 112
Table 3.9. Fraction of Drug Resistance to ESKAPE Organisms Attributable to
Antibiotic Use in ICU Patients, 1995-2002 ............................................. 113
Supplementary Table 3.1. Admitting Diagnoses assessed for ICU Patients .... 124
Supplementary Table 3.2. Select Comorbidities assessed for ICU Patients .... 129
Supplementary Table 3.3. Classification of Antibiotics Eliminated through the
Kidneys or through the Liver .................................................................. 130
Supplementary Table 3.4. Sensitivity Analysis for the Association between
Antibiotic Elimination Route and Drug Resistance of HA-BSI Organisms in
ICU Patients using Multinomial Model, 1995-2002 ................................ 132
Supplementary Table 3.5. Sensitivity Analysis for the Association between
Antibiotic Elimination Route and Drug Resistance of HA-BSI Organisms in
ICU Patients using Cumulative Model, 1995-2002 ................................ 133
Supplementary Table 3.6. Association between Type of Antibiotic as an Indicator
for Increasing Gut Exposure to Antibiotic Use and Drug Resistance of HA-
BSI Organisms in ICU Patients, 1995-2002 ........................................... 134
vii

Supplementary Table 3.7. Sensitivity Analysis for the Association between Type
of Antibiotic as an Indicator for Increasing Gut Exposure to Antibiotic Use
and Drug Resistance of HA-BSI Organisms in ICU Patients, 1995-2002
............................................................................................................... 135
Supplementary Table 3.8. Sensitivity Analysis for the Association between
Antibiotic Elimination Route and Drug Resistance of HA-BSI Organisms
Among ICU Patients with Antibiotic Use, 1995-2002 ............................. 136
Supplementary Table 3.9. Association between Antibiotic Elimination Route and
Drug Resistance of HA-BSI Organisms in ICU Patients, by Race, 1995-
2002 ....................................................................................................... 137
Supplementary Table 3.10. Association between Antibiotic Elimination Route and
Drug Resistance of HA-BSI Organisms in ICU Patients, by Sex, 1995-
2002 ....................................................................................................... 139
Supplementary Table 3.11. Association between Antibiotic Elimination Route and
Drug Resistance of HA-BSI Organisms in ICU Patients with a Positive
Blood Culture, by Gram Stain Classification, 1995-2002 ....................... 141
Supplementary Table 3.12. Association between Antibiotic Elimination Route and
Drug Resistance of HA-BSI Organisms among ICU Patients with Antibiotic
Use and Positive Blood Culture, 1995-2002 .......................................... 143
Supplementary Table 3.13. Sensitivity Analysis for the Fraction of Drug
Resistance Attributable to Antibiotic Use in ICU Patients, 1995-2002 ... 144
Table 4.1. Baseline and Demographic Characteristics of ICU Patients at Los
Angeles County Hospital (1995-2002) who were Included in the Study 161
Table 4.2. Association between Antibiotic Elimination Route and Degree of
Resistant Bacteremia in Participating patients, 1995-2002 .................... 164
Table 4.3. Sensitivity Analysis of the Association between Antibiotic Elimination
Route and Degree of Resistant Bacteremia in Participating Patients, by
Time from Antibiotic Start Date, 1995-2002 ........................................... 165
Supplementary Table 4.1. Classification of Antibiotics Eliminated through the
Kidneys or through the Liver .................................................................. 173
Supplementary Table 4.2. Association between Antibiotic Elimination Route and
Degree of Resistant Bacteremia in Participating Patients, by Select
Stratification Covariates, 1995-2002 ...................................................... 175

viii

List of Figures
Figure 1.1. Abundant bacterial phyla in healthy humans .................................... 25
Figure 1.2. Shift in Bacterial Diversity after Antibiotic Administration .................. 26
Figure 2.1. Process Flow of Analytic Datasets ................................................... 67
Figure 2.2. Bridging of hospitalization stays for contiguous and overlapping
records ..................................................................................................... 68
Figure 3.3. Minimum hospital stay and latency period required for qualifying
organisms ................................................................................................ 69
Figure 3.1. Patient Flowchart ............................................................................ 114
Figure 3.2. Distribution of antibiotics among ICU patients receiving antibiotics
eliminated through the kidneys or liver................................................... 115
Figure 3.3. Distribution of Select Antibiotics over Time among ICU Patients,
1995-2002 .............................................................................................. 116
Figure 3.4. Distribution of gram negative and gram positive organisms among
ICU patients with HA-BSI organisms ..................................................... 117
Figure 3.5. Distribution of Select Organisms over Time among ICU Patients with
a Positive Blood Culture, 1995-2002 ..................................................... 118
Figure 3.6. Distribution of Organism Resistance among Patients with a Positive
Blood Culture, by Admitting Year ........................................................... 119
Figure 4.1. Patient Flowchart ............................................................................ 166
Figure 4.2. Distribution of Organisms by Resistance Score ............................. 167
Figure 4.3. Survival probability of patients with positive blood cultures and prior
antibiotic use, stratified by antibiotic elimination route ........................... 168
Supplementary Figure 4.1. Time to blood culture among patients with positive
blood culture and prior antibiotic use, stratified by antibiotic elimination
route....................................................................................................... 177

ix

Abbreviations
AF Attributable fraction
ASP Antibiotic stewardship program
AR Antibiotic resistance
BSI Bloodstream infection
CDC Centers for Disease Control and Prevention
CMS Centers for Medicare and Medicaid Services
CRE Carbapenem resistant Enterobacteriaceae
ESKAPE Enterococcus faecium, Staphylococcus aureus, Klebsiella
pneumoniae, Acinetobacter baumannii, Pseudomonas
aeruginosa, Enterobacter species
ESBL Extended spectrum β-lactamase
FDA Food and Drug Administration
HA-BSI Hospital acquired bloodstream infection
HMP Human Microbiome Project
IBD Inflammatory bowel disease
ICU Intensive care unit
IDSA Infectious Disease Society of America
LAC-USC MC Los Angeles County University of Southern California Medical
Center
LOS Length of stay
MRSA Methicillin resistant Staphylococcus aureus
MSSA Methicillin susceptible Staphylococcus aureus
MSL Macrolides, licosamides, and streptogramins
NIH National Institute of Health
NNIS National Nosocomial Infections Surveillance
x

NOD Non-obese diabetic
PAF Population attributable fraction
PBP Penicillin binding protein
SSI Surgical site infection
UTI Urinary tract infection
VRE Vancomycin resistant Enterococci
WHO World Health Organization

xi

Abstract
Antibiotic use has been implicated in development of antibiotic resistance
by microbiological, animal, and epidemiological research. However, whether
antibiotics eliminated through the microbe-rich gut may confer greater risk of
resistance than antibiotics eliminated through the kidneys has not been
previously investigated. We addressed this novel hypothesis using data from
patients admitted to the Intensive Care Unit (ICU) of the Los Angeles County +
University of Southern California Medical Center (LAC+USC MC) between 1993
and 2003. Specifically, we collected data describing antibiotic use and antibiotic
resistance of organisms cultured from blood of patients with hospital-acquired
bloodstream infections (HA-BSI) and antibiotic administration records of patients
admitted to the ICU. We combined detailed demographic and treatment data on
the ICU patients to create two analytic cohorts: a health and policy cohort
representative of ICU patients (described in Chapters 2 and 3) and a clinically
relevant cohort of patients among whom HA-BSI was diagnosed following
administration of antibiotics (described in Chapter 4). Analyses of the public
health cohort were designed to describe the cohort, to understand trends of
antibiotic use and organisms identified from blood cultures among all patients
and those from whom ESKAPE organisms were cultured, and to estimate both
associations describing antibiotic-organism specific resistance and antibiotic
exposure-attributable fractions. Analyses of the clinically relevant cohort
addressed whether route of elimination of antibiotics was associated with
subsequent drug resistance and mortality.
xii

In Chapter 2, we describe sources of data received from the LAC+USC for
the study, as well as definitions, assumptions, and procedures that were used to
generate the overall cohort of ICU patients. Briefly, data were extracted from
clinical records and loaded as Microsoft ACCESS databases; subsequently
converted into Microsoft Excel Workbooks for clinical review; and imported into
SAS as sas7bdat files for data cleaning, manipulation, and assembly. The
independent variable, route of elimination of each antibiotic administered, was
classified using medication administration data from patients in the ICU. The
primary dependent variable was drug resistance of bacteria identified from blood
cultures which was based on antibiotic susceptibility assessment of organisms.
Additional outcome variables, including death, as well as demographic
characteristics, and select comorbidities were derived for all patients in the ICU.
The final analytic dataset included 11,576 patients, of whom 9,541 were
administered at least one antibiotic and 2,035 had no antibiotic use. A total of
10,497 patients had no eligible blood cultures. Of those with a positive blood
culture (N=1,079), 566 were non-resistant, 398 multi-drug resistant, and 115
extensive-drug resistant.
In Chapter 3 we report on analyses of the public health and policy cohort,
wherein we estimated associations between route of antibiotic elimination as well
as resistance and proportion of drug resistant HA-BSI organisms attributable to
each antibiotic type. We assessed drug resistance (no resistance, multi-drug
resistance, extensive-drug resistance) of organisms identified in subsequent HA-
BSI in relation to route of elimination of antibiotic administered (none, kidneys,
xiii

liver). Compared to patients administered no antibiotic, the odds of developing
multi-drug resistance was 1.82(95% CI: 1.11-2.97, p=0.0169)- versus 2.99(95%
CI: 1.84-4.85, p<0.0001)-times higher in those administered antibiotics eliminated
through the kidneys and liver, respectively; odds of developing extensive-drug
resistance was 3.46(95% CI: 0.82-14.68, p=0.0923)- versus 7.60(95% CI: 1.84-
31.41, p=0.0051)-times higher in these groups. Moreover, estimated resistance
attributable to antibiotics was 41.7% versus 72.8% among patients who received
antibiotics eliminated through the kidneys and liver, respectively.
In Chapter 4, we describe an extreme phenotype case-case study
(N=693) nested within the retrospective cohort described above (N=11,563).
Patients with a HA-BSI who were treated with at least one antibiotic were
classified according to the route of antibiotic elimination (liver or kidneys) and
according to drug resistance (none, multi-drug, or extensive-drug resistance). We
assessed the association between the route of elimination and both resistance
and mortality using multinomial regression and Cox proportional hazards model,
respectively. Patients administered any antibiotic eliminated through the liver had
increased odds of a subsequent drug-resistant BSI of any kind (OR=1.44,
p=0.0390) and increased risk of death (HRR=1.86, p=0.007).
We identified a pattern of drug resistance associated with antibiotic use, in
both the overall cohort of ICU patients (N=11,576) and a clinically relevant cohort
of patients stringently selected for antibiotic use and positive blood cultures
(N=693). Both multi-drug and extensive-drug resistance were more common in
patients who had been administered any form of antibiotics, but associations had
xiv

greatest magnitude for patients with antibiotics eliminated at least partially
through the liver.

1

Chapter 1: Background

2

1.1 INTRODUCTION
Hospital-acquired infections are a major health concern and are
associated with high mortality. However, many hospital-acquired infections are
difficult to treat due to resistance against many or all available antibiotics. There
has been an alarming increase in resistance rates of hospital-acquired infections
over several decades in the United States (US) and thus, efforts to fight the
antibiotic resistance problem have taken a multilevel approach. At the drug level,
efforts focus on the identification of antibiotics with novel mechanisms of action to
combat bacterial infections. At the hospital level, various strategies such as
antibiotic stewardship programs are implemented to minimize the emergence of
resistance and optimize antibiotic use and reduce antibiotic resistance.
Over the last decade microbiological advances have identified the
gastrointestinal tract, an environment consisting of immune cells, bacteria, and
foreign antigens, as a modulator of host health and immunity. Diseases such as
Type II diabetes, obesity, Crohn’s disease, inflammatory bowel disease, and
ulcerative colitis have been shown to be linked to disruption of the ecological
balance of the gut. The gut environment has also been shown to house
resistance machinery: the commensal microbiome that colonize the
gastrointestinal tract can confer antibiotic resistance owing to innate resistance
mechanisms and can provide an environment for the rapid exchange of genetic
material between bacterial strains and species. While microbiological and
epidemiologic studies have demonstrated that antibiotic use results in a greater
resistance profile of gut bacteria, no studies have explored potential modifiers of

3

the degree of antibiotic resistance. Here, we probe the link between degree of
antibiotic exposure in the gut, quantified by antibiotic elimination in the gut, and
degree of antibiotic resistance, categorized as no resistance, multi-drug
resistance, extensive-drug resistance, or pan-drug resistance. This dissertation
proposes two projects to identify the determinants of degree of antibiotic
resistance using a large hospital-based dataset. The first project will characterize
patterns of antibiotic use in organisms cultured from hospital-acquired
bloodstream infections (HA-BSIs) of patients in the ICU. We will determine if
antibiotic use is associated with drug resistance across groups of patients in the
ICU and determine the fraction of resistance attributable to antibiotic use among
patients in the ICU cohort. The second project will assess the association
between the route of antibiotic elimination and subsequent drug resistance and
survival among patients with antibiotic use and a positive blood culture.
1.2 HOSPITAL-ACQUIRED INFECTIONS
Hospital-acquired infections are a major cause of morbidity and mortality
in the US. Frequent types of hospital-acquired infections are urinary tract
infections (UTI), bloodstream infections (BSI), pneumonia, and surgical site
infections (SSI) [1]. Among these, BSIs can be particularly life-threatening and
are defined as the presence of a bacteria in the blood. BSIs were the 8
th
-12
th

leading cause of death in the US in 2014 in specific groups defined by age,
race/ethnicity, and sex [2]. The true incidence of BSIs is unknown, but estimates
range from approximately 535,920 to 628,320 episodes per year in the US [1, 3].
Patients with BSIs have a higher mortality rate and a longer length of hospital

4

stay compared to patients without BSIs (49% versus 33%, 23 days versus 15
days, respectively). Patients with BSIs also have twice the risk of dying within 90
days of infection compared to those without BSIs (OR: 2.08 (95% CI: 1.69, 2.57))
[4]. Patients admitted to the Intensive Care Unit (ICU) are a special population
within the hospital among whom infections can be readily transmitted. Such
patients are at an increased risk of developing hospital-acquired infections
because of the severity of their illness, prolonged hospital stay, use of infection-
associated devices, and surgical procedures [5]. It is estimated that half of all
HA-BSIs occur in ICU patients, and 26-48% of these result in death [6].
Antibiotic resistance is rampant among many of the common causative
organisms of hospital-acquired infections, and is a major world public health
concern [7]. Environmental-induced resistance to antibiotics is rising because of
antibiotic overprescribing and lack of compliance in following prescribed
antibiotics regimens. This has led to growing concern regarding the ability to
adequately treat hospital-acquired infections, and is compounded by approval of
few new drugs to treat infections. Both the Infectious Disease Society of America
(IDSA) and the World Health Association (WHO) warn that the pool of existing
and new antibiotics with which to treat patients is becoming increasingly small [7,
8]. The number of new antibiotics approved has decreased steadily over the past
35 years, and only 8 antibiotics were approved between 2010 and 2015. Of
these, only 3 have in vitro activity against the most resistant pathogens,
Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae,
Acinetobacter baumannii, Pseudomonas aeruginosa, Enterobacter species

5

(ESKAPE). Of the 7 non-tuberculosis antibiotics, none have a new mechanism of
action [9]. The impact of increasing resistance to antibiotics is far-reaching and
presents both a health and economic burden due to longer length of hospital
stay, extra patient care, and higher cost of non-resistant antibiotics.
Strategies to Minimize Hospital-Acquired Infections
Several strategies have been implemented in efforts to reduce hospital
dissemination of resistant bacteria. Interventions based on preventing and
minimizing transmission of bacteria include hygiene (hand-washing, changing
gloves and gowns), reducing length of stay in the hospital, and appropriate use
and monitoring of catheters. Other interventions, such as antibiotic cycling and
antibiotic restriction policies through Antibiotic Stewardship, are based on
preventing the emergence of antibiotic resistance [10]. Antibiotic cycling is the
use of different classes of antibiotics with similar spectrum of activity but different
mechanisms of action for limited periods of time. The theoretical basis of this is
that rotating antibiotics with different mechanisms of action can limit the
emergence of resistant bacteria; however, studies which assess the
effectiveness of antibiotic cycling have mixed results [11]. While some studies
have found increased susceptibility of bacteria to specific antibiotics with this
intervention, others demonstrate no substantial change in susceptibility [10].
An Antibiotic Stewardship Program (ASP) places a restriction policy on
antibiotics associated with resistance to monitor and improve antimicrobial
prescribing practices, whereby physician requests for restricted antibiotics must
be approved by the stewardship team. Approval is based on whether the
specified antibiotic is appropriate for the patient indication. A restriction policy is

6

implemented institution-wide and is a coordinated effort between physicians and
pharmacy; therefore antibiotics on the restricted list cannot be dispensed without
approval. The goals are to limit the use of broad spectrum antibiotics and to
reduce the high cost of antibiotics [12]. Data from several studies indicate that
the ASP interventions are effective in limiting antibiotic prescriptions. A cluster
randomized clinical trial at 47 primary care facilities found that the ASP
intervention decreased inappropriate antibiotic prescribing rates significantly for
patients with acute respiratory tract infections [13]. However, it has been difficult
to establish the effectiveness of this intervention in minimizing the emergence of
resistant infections. For example, one study found that although the ASP had
been in place at a hospital since 2001, a significant decrease in the proportion of
MRSA and Clostridium difficile infections in the hospital was only seen after the
implementation of an electronic medical record (EMR) database to facilitate the
ASP in decision-making [14].
In summary, antibiotic cycling and antibiotic restriction are two strategies
aimed at limiting and preventing the emergence of antimicrobial resistance,
although there are mixed results on the ability of these interventions to achieve
their goals. These strategies are not sufficient in preventing antibiotic resistance
emergence.
1.3 CHARACTERISTICS OF BACTERIA IN THE HOSPITAL
An 8 year prospective nationwide study of 49 hospitals across the United
States found that the majority (65%) of single organisms identified from BSIs
were gram-positive. Major pathogens found in the hospital setting were

7

coagulase-negative staphylococci (31%), Staphylococcus aureus (20%),
Enterococcus species (9%), Candida species (9%), Escherichia coli (6%), and
Klebsiella species (5%) [6]. In general, the most resistant bacteria identified in
the hospital setting are Enterococcus faecium, Staphylococcus aureus, Klebsiella
pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and
Enterobacter species, collectively referred to as the ESKAPE bacteria [15, 16].
ESKAPE pathogens exhibit high resistance rates that have increased
significantly over the decades. These bacteria can be resistant against a single
class of drugs or against multiple classes of drugs and pose a threat to our ability
to treat hospital-acquired infections. Data from the National Nosocomial
Infections Surveillance (NNIS) System show an approximately 60% increase,
25% increase, and 30% increase in strains of methicillin-resistant
Staphylococcus aureus (MRSA), vancomycin-resistant Enterococcus (VRE), and
fluoroquinolone-resistant Pseudomonas aeruginosa respectively between the
1980s and 2002 [1]. The Centers for Disease Control and Prevention (CDC)
describes the drug-resistant ESKAPE pathogens as a serious public health
threat. The exception is extended spectrum β-lactamase producing (ESBL)
Enterobacteriaceae which is classified as an urgent public health threat [17].
Approximately 140,000 Enterobacteriaceae infections in the hospital occur
each year, of which an estimated 9,300 (11%) are carbapenem-resistant
Enterobacteriaceae (CRE), and of these, 88% are carbapenem-resistant
Klebsiella species [17]. Carbapenem-resistant strains have become resistant
against all available antibiotics and are on the rise in hospitals. Extended-

8

spectrum β-lactamase producing Enterobacteriaceae are multi-drug resistant
infections that pose a problem because these organisms are resistant to the first
line of therapy (penicillins and cephalosporins). Therefore patients with these
infections can be prescribed carbapenems, contributing to an increase in CRE.
Although there are fewer Acinetobacter infections in the hospital per year
(approximately 12,000), an estimated 63% of these infections are multi-drug
resistant and are therefore becoming more difficult to treat. One of the most
common causes of hospital-acquired infections is Staphylococcus aureus, and
specifically methicillin resistant Staphylococcus aureus (MRSA). Rates of
invasive MRSA in the hospital decreased from 2005-2011, primarily due to
improved hygiene and procedures associated with catheter-lines [17]. Although
the annual number of MRSA infections is unknown, it is estimated that 80,461
invasive MRSA infections and 11,285 related deaths occur per year. The CDC
estimates 66,000 Enterococcus infections per year. Approximately one-third of
these infections are vancomycin-resistant and 1,300 patient deaths are
attributable to drug resistant Enterococcus per year. Of the Enterococci,
Enterococcus faecium exhibits the highest resistance to vancomycin at 77% [17].
Although vancomycin had been available for several decades, vancomycin
resistant Enterococci surfaced much later in 1987 [17, 18].
1.4 THE GUT MICROBIOME AND HUMAN HEALTH
The lumen of the intestines in the gastrointestinal tract is a milieu of
commensal bacteria, food antigens, pathogens, antimicrobial peptides, and cells
of immune response (immunoglobulin A, effector T cells, regulatory T cells, B

9

cells, neutrophils, and interleukins) that together regulate nutrition, mucosal and
non-mucosal host immunity and health. The gut maintains a fine balance of
immunity and must remain toleragenic in the presence of food and nutrition but
have the ability to generate an inflammatory response in the presence of
pathogen infiltration [19]. In general, the toleragenic response is modulated by
regulatory T cells which suppress an effector T cell response to create tolerance.
The inflammatory response is promoted by Th17 cells, which are a subset of
CD4 T helper cells that are distinct from CD4 TH1 cells that trigger an
inflammatory response. They secrete IL-17 and promote IgA production in the
presence of pathogens or toxins. Although they trigger an inflammatory response
that is milder than the TH1 inflammatory response, they are nonetheless
associated with autoimmune and inflammatory diseases [19].
Commensal bacteria in the gut play a vital role in regulating a toleragenic
or inflammatory response, such as by inducing a Th17 response against specific
pathogens. Studies using animal models demonstrate a causal relationship
between bacterial distribution of the gut, gut health, and disease states. For
example, germ-free mice with no previous obesity can be induced to become
obese by transplanting the microbiome from obese mice into germ-free mice.
Fecal transplant studies also show that transplanting germ-free mice with the
microbiome of mice with induced colitis results in the development of colitis in
germ-free mice [19]. Disturbances in the gut microbiome can directly impact the
type of immune response. In one study, mice colonized with the commensal
segmented filamentous bacteria in the gut had upregulation of elements that led

10

to a Th17 pro-inflammatory response. These mice also had a protected mucosal
barrier that prevented the penetration of Citrobater rodentium. Mice not colonized
with the segmented filamentous bacteria had no Th17 response and had higher
invasion of Citrobacter rodentium, indicating that commensal bacteria can directly
induce an immune response as well as provide a protective barrier against
invasion [20]. Animal models have shown that antibiotic administration can
induce an immune response through disturbances or ablation of the commensal
gut microbiomes, thereby changing the functional profile of the gut microbiome.
One study found that after vancomycin administration, young non-obese diabetic
(NOD) mice that spontaneously develop Type I diabetes exhibited substantially
lower levels of IL-17, a pro-inflammatory interleukin, but a higher proportion of
FoxP3+, a regulatory cell that blocks the TH1 inflammatory response, in the gut
compared to control mice [21].
Until recently, characterizing the genetic diversity of the bacteria in the
gastrointestinal tract was limited to organisms identified from traditional culture-
based techniques. In 2008 the National Institute of Health (NIH) led The Human
Microbiome Project (HMP) initiative with the goal of characterizing the human
microbiome at a variety of sites including nasal and oral cavities, lung,
gastrointestinal tract and urogenital tract using whole genome shotgun
sequencing and 16S rRNA gene sequencing. Interrogation of the genetic
makeup of the gastrointestinal tract revealed 70-100 known bacterial species
using whole genome sequencing and an estimated 1,000 bacterial species using
16S rRNA gene based sequencing [22-24]. Although there is substantial intra-

11

person and inter-person diversity of the gut microbiome, a core set of functional
bacteria have been identified as being necessary for human health, from which a
departure can result in altered immune function and disease [23, 25, 26]. A
healthy gut is characterized as predominantly having the bacterial phyla
Firmicutes and Bacteroidetes and the taxa Bacteroides, Bifidobacterium,
Faecalibacterium, Eubacterium, Roseburia, and Lactobacillus (Figure 1, adapted
from Ref [27]) [24, 27, 28]. Studies that examine the relationship between
gastrointestinal bacteria and human health are numerous and illustrate a clear
role of the richness, diversity, and evenness of bacteria in the gastrointestinal
tract as a modulator of immunity and diseases [24, 26, 29, 30].
Dysbiosis, a disruption of the gut microbiome, can have profound
consequences on the host health. It is associated with a myriad of disease
including Type II diabetes, ulcerative colitis, inflammatory bowel disease (IBD),
asthma and allergy, and obesity. A shift in the core bacteria results in either
upregulation or downregulation of cells that directly mediate the immune
response. For example, some allergic diseases are triggered by an anti-
inflammatory response from the gut. Autoimmune disorders such Crohn’s
disease are caused by an increased pro-inflammatory response due to higher
Th2 expression in the gut and, in IBD, by higher Treg expression in the gut [26].
Characterization of the gut microbiome of healthy and unhealthy persons indicate
that the distribution of bacteria phyla are markedly different between the two
groups. Humans with IBD have a greater distribution of Proteobacteria and
Acinitobacteria and a much lower distribution of Bacteroidetes and Firmicutes in

12

their gut compared to heathy humans without IBD. Humans with Type II diabetes
harbor a greater distribution of Bacteroidetes and less Firmicutes in the gut
compared to humans without Type II diabetes [27]. Thus both the presence and
distribution of specific bacterial phyla and taxa are important for proper
gastrointestinal function and can determine disease states.
1.5 MECHANISMS OF ANTIBIOTIC RESISTANCE
Antibiotic resistance is the continued growth of bacteria in the presence of
a cytotoxic or cytostatic antibiotic. Bacterial mechanisms of resistance can be
intrinsic or extrinsic. Intrinsic resistance is based on the natural resistance of an
organism in the absence of environmental exposure to the antibiotic. Two such
examples are the 30,000 year-old soil bacterial samples identified from the Late
Pleistocene period from Dawson City, Yukon, Canada and separately, the over 4
million year-old soil bacteria identified from caves in Carlsbad Caverns National
Park, New Mexico. Both of these bacterial samples had antibacterial-resistance
genes and demonstrated multi-drug resistance against modern antibiotics [31,
32]. Extrinsic resistance is the acquired resistance of bacteria to antibiotics
against which they previously susceptible. A major contributor to the
development of resistance is frequency and duration of antibiotic exposure,
which provides selective pressure [33]. Bacteria can acquire resistance after
exposure to antibiotics owing to factors such as the large quantity of bacteria
present at a site, the rapid reproduction time, and mutation rate. Combining these
factors with exposure to antibiotics eventually leads to mutated bacteria with
antibiotic resistance. Emergence of resistance arises when, at a site, exposure to

13

an antibiotic kills susceptible bacteria and selects resistant bacteria to reproduce
and become the dominant bacteria (Figure 2, adapted from Ref [34]) [18].
Resistance can then be easily transferred with the exchange of the resistant
bacterial genome, based on gene transfer by conjugation, transformation, or
transduction [33]. Bacterial resistance mechanisms can be general or specific to
antibiotic classes and compounds.
The three main bacterial processes targeted by antibiotics, cell wall
biosynthesis, protein synthesis, and DNA replication/repair, can be counteracted
by mechanisms expressed in bacteria. These mechanisms are the antibiotics
efflux pumps, enzymatic modification or deactivation of antibiotics, and alteration
of bacterial targets [18]. Efflux pumps in the cell membrane are ubiquitous and
critical in establishing chemical gradients, such as in ion transport or the export of
toxic compounds. Efflux-mediated antibiotic resistance involves the
overexpression of genes which encode efflux pumps on the bacterial cell
membrane, resulting in sub-therapeutic concentrations of intra-bacterial
antibiotics and thereby rendering the antibiotic ineffective [18]. Efflux-mediated
resistance has been documented in both gram-positive and gram-negative
bacteria. For example, Pseudomonas aeruginosa has efflux pumps which confer
intrinsic resistance non-selectively to tetracyclines, chloramphenicol, norfloxacin,
linezolid, glycylcycline, and tigecycline. Additional examples are the efflux-
mediated resistance of Escherichia coli to linezolid and macrolides and
Haemophilus influenza to macrolides [35].

14

Antibiotic deactivation generally involves cleaving of or binding to the
antibiotic to inactive the compound or reduce its affinity to the bacterial target.
For example, the β-lactam-containing penicillins and cephalosporins, which
target bacterial penicillin-binding proteins, can be inactivated by the β-lactamase
enzyme. This enzyme cleaves the β-lactam ring of penicillins and
cephalosporins, resulting in inactive antibiotics [5, 18]. The Enterobacteriaceae
family, which includes the Enterobacter and Klebsiella genera, has the ability to
produce narrow- and extended-spectrum β-lactamase [5]. The most common
variants of the narrow-spectrum β-lactamase are TEM-1 and SHV-1, while ESBL
has a few common variants of TEM and SHV. Transmission of ESBL occurs via
plasmids, which often carry resistant genes for aminoglycosides, trimethoprim-
sulfamethoxazole, and tetracycline [5]. Protein-synthesis inhibitors such as
aminoglycosides work by binding to RNA and interrupting protein synthesis in the
ribosome. Bacterial enzymes can bind to the aminogylcosides before these
antibiotics reach the ribosome, and lower the antibiotic binding affinity to RNA to
prevent interruption of protein synthesis [18]. Bacteria can have multiple resistant
mechanisms that work in combination to have a joint effect against an antibiotic.
For example, the bacteria Pseudomonas aeruginosa has both efflux pumps
which prevent an effective antibiotic concentration inside the cell and enzymes
which deactivate antibiotics.
Another mechanism of resistance involves altering the bacterial target of
antibiotics that interrupt protein synthesis. A variety of bacterial genes with this
mechanism have been identified. The Erm methyltransferase enzyme modifies

15

the 23S RNA component of ribosomes and reduces the antibiotic’s ability to
interrupt protein synthesis, and is the primary resistant mechanism of
Staphylococcus aureus. The vanHAX genes encoded in Enterococci use a
similar mechanism to significantly reduce vancomycin’s binding affinity to
peptidoglycans of the bacterial cell wall. Bacteria can also have mutations of the
target proteins. For example, the mecA gene in Staphylococcus aureus encodes
a mutated penicillin-binding protein (PBP2a) that lowers the affinity of β-lactam
antibiotics and allows its escape from methicillin [5, 18]. Transmission of this
resistant element is low, and therefore it is suspected that the predominant
manner of MRSA spread is by transmission of organism [5].
1.6 THE ANTIBIOTIC RESISTOME IN HUMANS
Gene sequencing techniques have identified a large number of resistant
genes encoded by commensal bacteria in the human gut. The antibiotic
resistome, a term coined by G.D. Wright in 2007, describes the set of all
antibiotic resistant genes present in bacterial chromosomes. It includes all
resistant genes, expressed and unexpressed genes, cryptic elements, and
resistant precursor genes that represent moderate resistance but have the
potential to develop severe resistance in the future [36]. Genomic studies have
identified that SNPs in the human gut microbiome most frequently occur in genes
associated with conferring bacterial resistance [25]. The Antibiotics Resistance
Genes Database, a compendium of resistant genes, has identified 23,137
resistance genes in the human microbiome that map to 380 distinct gene clusters

16

with the same resistance mechanism and resistance profile and are resistant to
249 known antibiotics [37].
The antibiotic resistome of bacteria in the human gut has been
characterized to compare its resistance profile with that of pathogenic bacteria
found in the environment. Out of the 4.1 million bacterial genes in the gut, 1,093
resistant genes have been identified belonging predominantly to the Firmicutes
(52%), Bacteroidetes (15%), and Proteobacteria (32%) phyla. The distribution of
antibiotic resistant genes is lower in the Bacteroidetes phylum and higher in the
Proteobacteria phylum compared to non-resistant gut genes. These resistant
genes mapped to 149 unique resistant gene types, of which 95 are single-drug
resistant and 54 are multi-drug resistant [23]. Bacterial genes confer single-drug
resistance most frequently against tetracycline, followed by bacitracin and then
vancomycin [38]. Among genes conferring multi-drug resistance, those conferring
resistance to the macrolides, licosamides, and streptogramins (MSLs) are the
most abundant [23, 38]. Overall, approximately 50% of the resistant genes in the
human gut are resistant to tetracyclines, and the majority (approximately 75%) of
resistance genes are resistant to tetracyclines, MSLs or beta-lactams [23].
Although there is country-specific variation in the distribution of resistant genes,
the tetracycline resistance gene type TetQ was the most abundant across
countries [23, 38]. Among other resistance gene type families, ErmB, ErmF, and
ErmG were the most common Erm genes, Bl2e-cfxa was the most common
cephalosporin resistance gene type, and BacA was the most common bacitracin
resistance gene type [23].

17

The antibiotic resistome is found in humans even in the absence of
antibiotic exposure, indicating that genetic exchange of resistant material and
selection pressure due to antibiotic exposure do not explain all of the resistant
bacteria and genes harbored inside the human gut. Fecal samples from a study
of uncontacted Amerindians removed from Western society and with no antibiotic
exposure for at least 11,000 years revealed bacteria containing functional
antibiotic resistance genes. This study found that the gut microbiome of the
Yanomami Amerindians contained greater bacterial diversity compared to the
Guahibo Amerindians, Malawians, and subjects residing in the Unites States. It
also found presence of antibiotic resistance genes that were resistant to both
synthetic antibiotics and recent third- and fourth-generation cephalosporins,
indicating that resistance genes of gut commensal bacteria may predate modern
antibiotics. Interestingly, these resistance genes are homologs to the resistance
genes from microbiomes collected in the Human Microbiome Project, identified in
persons from countries of antibiotic use [39]. Another study examining the gut
microbiome of 16 infants found the presence of antibiotic resistant bacteria and
antibiotic resistance genes within the first week after birth. The Tet resistant
bacteria and tet(M) gene pool was prevalent in the stools of newborn infants and
remained at a consistent level one year after birth. Erm resistant bacteria were
also found early within days after birth, and the count of Erm resistant bacteria
and ErmB gene pool rose until reaching a plateau. Antibiotic resistant bacteria of
the infant gut closely resembled the skin bacteria of mothers. One possible
explanation for these findings is that natural birth confers antibiotic resistant

18

bacteria through maternal and environmental transmission, and that amplification
of resistance genes can occur with age due to lack of selective pressure [40].
The antibiotic resistome in the human gut provides a pool of genetic
information readily available for intra-species and inter-species bacterial transfer.
An illustration of this is the Bacteroidales strains which exchange 140kB of DNA
within species and also between species [41]. Mobile elements and horizontal
gene transfer are key mechanisms of genetic transfer. Resistant bacteria
selected preferentially after selective pressure from antibiotic exposure will
transfer resistant genes from the antibiotic resistant gene pool.
1.7 ANTIBIOTIC USE PERTURBS THE BACTERIAL DISTRIBUTION OF THE
GUT AND PROMOTES EMERGENCE OF RESISTANCE
Commensal bacteria that colonize the gastrointestinal tract are adversely
affected by antibiotic exposure. Microbiological studies show that changes in the
gut bacterial environment are rapid and persistent after antibiotic exposure [34,
42-45]. Antibiotic use decreases diversity of core bacteria in the gut both during
and after antibiotic administration, with partial recovery of bacteria and
emergence of resistant bacteria and resistant genes [43]. One study which
followed short-term administration of metronidazole, clarithromycin, and
omeprazole in patients found a change in the bacterial phyla distribution of the
fecal samples immediately after gut exposure to antibiotics. This change in
bacterial diversity persisted for 4 years after the initial antibiotic administration
[42]. Another study revealed persistently elevated levels of antibiotic resistance
genes for 2 years after initial antibiotic exposure, among patients exposed to

19

antibiotics [44]. While the gut can recover to some degree with single antibiotic
administration, repeated exposure to antibiotics decreases genetic diversity. In
Dethlefsen et al., a short course of ciprofloxacin administered in a group of
healthy subjects resulted in decreased richness, diversity, and evenness
immediately after administration but stabilized approximately 4 months after
administration; however, repeated administration of antibiotics in subjects
resulted in diminished the richness and phylogenetic diversity of the gut
environment [45, 46].
The presence of antibiotics reduces the gut’s ability to resist colonization
of dominating and potentially pathogenic bacteria, and leads to the emergence of
antibiotic resistance due to selective pressure. In Jakobsson et al., antibiotic
resistance genes that were not detected in patients at baseline were detected at
high levels immediately after treatment and persisted for 1 year and 4 years after
first administration of the antibiotics [42]. A separate 7-day study of clindamycin
administration in patients found that resistance to erythromycin, clindamycin, and
tetracycline increased to between 70-90% compared to baseline. Resistance
decreased over time but remained elevated after 2 years compared to baseline
resistance levels. ErmB resistant genes were found elevated up to 6 months after
antibiotic administration [47]. Patients with prior antibiotic use are shown to have
increased resistance to antibiotics compared to patients with no prior antibiotic
use. Nyberg, et al. showed that patients with prior clindamycin use had increased
resistance to ampicillin, sulfamethoxazole, chloramphenicol, streptomycin, and

20

trimethoprim for one year while patients with no prior clindamycin use showed no
increased resistance to these antibiotics [48].
1.8 ANTIBIOTIC METABOLISM AND ELIMINATION
In metabolism, the primary goal is to convert lipid-soluble xenobiotic
compounds such as antibiotics into polar compounds that can be eliminated.
Blood carrying compounds such as nutrients and drugs is metabolized by series
of phase I and phase II reactions. Phase I and II reactions can occur in both the
kidney via glomerular filtration or via the liver and biliary tree, although the
highest levels of xenobiotic-metabolizing enzymes are in the gastrointestinal
tract. Phase I reactions introduce a functional group such as a hydroxyl group (-
OH) to the xenobiotic compound, which often inactivates the pharmacological
activity of the compound. Phase II conjugation reactions combine endogenously-
derived charged and water-soluble species, glucuronic acid, sulphate,
glutathione, amino acids, or acetate, covalently to the xenobiotic metabolites
produced from Phase I reactions. The resulting polar compounds, which are
typically inactive, have increased solubility and can be excreted through the urine
or stool.
Xenobiotics can be eliminated unchanged or as metabolites of the parent
drug. Organs of excretion are the kidneys, liver, lungs, and skin, although the
kidney and liver are the major sites of elimination. The kidneys are able to
excrete small hydrophilic molecules. Larger hydrophilic molecules, hydrophobic
compounds, and molecules bound to plasma proteins such as many antibiotics
are eliminated hepatically by passing through the liver and the bile [33]. These

21

compounds enter the liver via the portal vein where they are detoxified by
hepatocytes, and subsequently released into the bile and gastrointestinal tract,
where they can either undergo reabsorption into blood (enterohepatic recycling)
or be eliminated via feces [49]. If the drug goes through enterohepatic recycling,
it will be passed through the liver and gastrointestinal tract again and the
unabsorbed drug will be eliminated via feces. Orally administered drugs are
generally absorbed rapidly in the gut, and go through first-pass metabolism and
enter the liver via the portal vein for a second pass through the gastrointestinal
tract, followed by elimination into feces. Intravenously administered antibiotics go
through systemic circulation and enter the liver through the portal vein, where
they can be eliminated through the gastrointestinal tract or be filtered in blood to
the kidneys [49]. In the kidneys, drugs go through glomerular filtration, active
tubular secretion, and passive tubular reabsorption resulting in elimination of the
compound in urine [33]. Often elimination of antibiotics involves a combination of
renal and hepatic elimination, although some antibiotics are eliminated
predominantly via the kidneys and others predominantly via the liver. Because
the renal excretion of antibiotics involves passing through a sterile environment
[50] and does not involve passing through the gastrointestinal tract, which has an
extensive bacterial community, antibiotic exposure in the gastrointestinal tract
can be determined largely by the antibiotic elimination route. The degree of
antibiotic exposure in the gastrointestinal tract affects the gut microbial
environment and leads to resistance gene pools and selective resistance niches.

22

Numerous antibiotic-specific pharmacokinetic studies have quantified drug
elimination by measuring parent drug and its metabolites in the urine and feces.
However, these estimates on a continuous scale are prone to large variability
and can be imprecise due to small sample size of healthy study participants and
substantial inter-person variability. Moreover, estimates generally focus on
antibiotic excretion in the urine because measurements of antibiotic in the bile
and feces are more difficult, intensive, and costly to obtain. A suitable alternative
criteria for determining the major route of antibiotic elimination is whether dose
adjustment of the antibiotic is required for a patient with renal impairment. Renal
impairment due to disease can adversely impact the body’s ability to renally clear
xenobiotics such as antibiotics. Presence of renal impairment is established
using creatinine clearance value. If an antibiotic is predominantly eliminated via
the kidneys, then dose adjustment will be required for patients with diminished
renal function. If an antibiotic is eliminated predominantly via the liver and biliary
tree, then no dose adjustment is required. Thus, whether or not an antibiotic
requires dose adjustment can be used to infer its major route of elimination.
The few studies that have examined the relationship between the
pharmacologic properties of antibiotics and antibiotic resistance, conducted
mainly in animal models [51, 52], have been limited to administration routes. One
study examining feces of mice inoculated with non-resistant Enterococcus
species or Escherichia coli found that the administration route of antibiotics was
associated with both the antibiotic resistance gene pool and presence of
antibiotic resistant bacteria [51]. This study found that orally administered

23

ampicillin conferred greater resistance compared to intravenously administered
ampicillin, and both orally and intravenously administered tetracycline conferred
greater resistance. This difference in resistance is likely due to the difference in
intravenous antibiotic exposure in the gastrointestinal tract: intravenous ampicillin
is excreted primarily renally whereas intravenous tetracycline is excreted both
hepatically and renally. Another study found that oral antibiotics conferred
increased resistance in chickens [52]. Both studies show evidence that oral
antibiotics confer antibiotic resistance, likely by way of antibiotic exposure to the
gut. The former study also shows evidence of antibiotic elimination route being a
modifier of resistance. This suggests an unexplored pathway to antibiotic
resistance.
To assess the associations between the route of antibiotic elimination and
drug resistance and its impact on other patient factors such as mortality, we
propose the specific aims, below.
1.9 SPECIFIC AIMS
Project 1
Aim 1: Characterize the population of ICU patients in the LAC-USC MC between
1995 and 2002 with respect to hospital characteristics, patient
characteristics, and comorbidities.
Aim 2: Describe patterns of HA-BSIs, resistance of HA-BSI organisms, and
antibiotic use among ICU patients at LAC-USC MC.
Aim 3: Determine if any form of antibiotic use is associated with resistance,
overall and according to the route of elimination, for all patients, patients

24

with antibiotic use, patients with antibiotic use and organisms, and patients
with ESKAPE organisms.
Aim 4: Determine whether age, sex, type of ICU, ICU length of stay, year of
admission, gram stain classification, race, and select baseline
comorbidities in the ICU confound the association or are effect modifiers
of the association between antibiotic elimination route and drug resistance
in HA-BSIs among patients in the LAC-USC MC ICU.
Aim 5: Determine the proportion of resistant HA-BSIs attributable to any antibiotic
use among patients in the LAC-USC MC ICU.
Project 2
Aim 1: Determine if the antibiotic elimination route is associated with subsequent
drug resistance in HA-BSIs among patients in the LAC-USC MC ICU with
antibiotic use and HA-BSI, after adjusting for sex, length of stay in the
ICU, and other health hospital related factors.
Aim 2: Determine if the antibiotic elimination route is associated with mortality in
the LAC-USC MC ICU, after adjusting for sex, length of stay in the ICU,
and other health and hospital related factors.
Aim 3: Determine whether age, sex, and gram stain classification are effect
modifiers of the association between antibiotic elimination route and drug
resistance in HA-BSIs among patients in the LAC-USC MC ICU with
antibiotic use and HA-BSI.

25

1.10 FIGURES
Figure 1.1. Abundant bacterial phyla in healthy humans

Adapted from Ref [27]

26

Figure 1.2. Shift in Bacterial Diversity after Antibiotic Administration

Adapted from Ref [34]

27

1.11 REFERENCES
1. National Nosocomial Infections Surveillance, National Nosocomial
Infections Surveillance (NNIS) System Report, data summary from
January 1992 through June 2003, issued August 2003. Am J Infect
Control, 2003. 31(8): p. 481-98.
2. Heron, M., Deaths: Leading Causes for 2014. Natl Vital Stat Rep, 2016.
65(5): p. 1-96.
3. Goto, M. and M.N. Al-Hasan, Overall burden of bloodstream infection and
nosocomial bloodstream infection in North America and Europe. Clin
Microbiol Infect, 2013. 19(6): p. 501-9.
4. Kaye, K.S., et al., Effect of nosocomial bloodstream infections on
mortality, length of stay, and hospital costs in older adults. J Am Geriatr
Soc, 2014. 62(2): p. 306-11.
5. Rice, L.B., Controlling antibiotic resistance in the ICU: different bacteria,
different strategies. Cleve Clin J Med, 2003. 70(9): p. 793-800.
6. Wisplinghoff, H., et al., Nosocomial bloodstream infections in US
hospitals: analysis of 24,179 cases from a prospective nationwide
surveillance study. Clin Infect Dis, 2004. 39(3): p. 309-17.
7. World Health Organization, Antimicrobial Resistance: Global Report on
Surveillance, 2014. World Health Organization.
8. Infectious Disease Society of America, Bad Bugs, No Drugs: as antibiotic
discovery stagnates, a public health crisis brews, 2004.

28

9. Deak, D., et al., Progress in the Fight Against Multidrug-Resistant
Bacteria? A Review of U.S. Food and Drug Administration-Approved
Antibiotics, 2010-2015. Ann Intern Med, 2016. 165(5): p. 363-72.
10. Kollef, M.H. and V.J. Fraser, Antibiotic resistance in the intensive care
unit. Ann Intern Med, 2001. 134(4): p. 298-314.
11. Lee, C.R., et al., Strategies to minimize antibiotic resistance. Int J Environ
Res Public Health, 2013. 10(9): p. 4274-305.
12. Centres for Disease Control and Prevention, Core Elements of Hospital
Antibiotic Stewardship Programs, 2014; Available from
http://www.cdc.gov/getsmart/healthcare/implementation/core-
elements.html.
13. Meeker, D., et al., Effect of Behavioral Interventions on Inappropriate
Antibiotic Prescribing Among Primary Care Practices: A Randomized
Clinical Trial. JAMA, 2016. 315(6): p. 562-70.
14. Cook, P.P., et al., Sustained reduction in antimicrobial use and decrease
in methicillin-resistant Staphylococcus aureus and Clostridium difficile
infections following implementation of an electronic medical record at a
tertiary-care teaching hospital. J Antimicrob Chemother, 2011. 66(1): p.
205-9.
15. Rice, L.B., Federal funding for the study of antimicrobial resistance in
nosocomial pathogens: no ESKAPE. J Infect Dis, 2008. 197(8): p. 1079-
81.

29

16. Boucher, H.W., et al., Bad bugs, no drugs: no ESKAPE! An update from
the Infectious Diseases Society of America. Clin Infect Dis, 2009. 48(1): p.
1-12.
17. Centres for Disease Control and Prevention, Antibiotic Resistance Threats
in the United States, 2013. Centres for Disease Control and Prevention,
US Department of Health and Human Services.
18. Walsh, C., Molecular mechanisms that confer antibacterial drug
resistance. Nature, 2000. 406(6797): p. 775-81.
19. DePaolo, R., Mucosal Immunity: a delicate balance between tolerance
and inflammation (Powerpoint slides). 2013.
20. Ivanov, II, et al., Induction of intestinal Th17 cells by segmented
filamentous bacteria. Cell, 2009. 139(3): p. 485-98.
21. Candon, S., et al., Antibiotics in early life alter the gut microbiome and
increase disease incidence in a spontaneous mouse model of
autoimmune insulin-dependent diabetes. PLoS One, 2015. 10(5): p.
e0125448.
22. Methe, B., et al., A framework for human microbiome research. Nature,
2012. 486(7402): p. 215-21.
23. Hu, Y., et al., Metagenome-wide analysis of antibiotic resistance genes in
a large cohort of human gut microbiota. Nat Commun, 2013. 4: p. 2151.
24. Hollister, E.B., C. Gao, and J. Versalovic, Compositional and functional
features of the gastrointestinal microbiome and their effects on human
health. Gastroenterology, 2014. 146(6): p. 1449-58.

30

25. Schloissnig, S., et al., Genomic variation landscape of the human gut
microbiome. Nature, 2013. 493(7430): p. 45-50.
26. Fujimura, K.E., et al., Role of the gut microbiota in defining human health.
Expert Rev Anti Infect Ther, 2010. 8(4): p. 435-54.
27. Spor, A., O. Koren, and R. Ley, Unravelling the effects of the environment
and host genotype on the gut microbiome. Nat Rev Microbiol, 2011. 9(4):
p. 279-90.
28. The Human Microbiome Project Consortium, Structure, function and
diversity of the healthy human microbiome. Nature, 2012. 486(7402): p.
207-14.
29. Shreiner, A.B., J.Y. Kao, and V.B. Young, The gut microbiome in health
and in disease. Curr Opin Gastroenterol, 2015. 31(1): p. 69-75.
30. Wu, G.D. and J.D. Lewis, Analysis of the human gut microbiome and
association with disease. Clin Gastroenterol Hepatol, 2013. 11(7): p. 774-
7.
31. D'Costa, V.M., et al., Antibiotic resistance is ancient. Nature, 2011.
477(7365): p. 457-61.
32. Bhullar, K., et al., Antibiotic resistance is prevalent in an isolated cave
microbiome. PLoS One, 2012. 7(4): p. e34953.
33. Greenwood, D., Medical Microbiology, 18th edition. 2012.
34. Jernberg, C., et al., Long-term impacts of antibiotic exposure on the
human intestinal microbiota. Microbiology, 2010. 156(Pt 11): p. 3216-23.

31

35. Cox, G. and G.D. Wright, Intrinsic antibiotic resistance: mechanisms,
origins, challenges and solutions. Int J Med Microbiol, 2013. 303(6-7): p.
287-92.
36. Wright, G.D., The antibiotic resistome: the nexus of chemical and genetic
diversity. Nat Rev Microbiol, 2007. 5(3): p. 175-86.
37. Liu, B. and M. Pop, ARDB--Antibiotic Resistance Genes Database.
Nucleic Acids Res, 2009. 37(Database issue): p. D443-7.
38. Ghosh, T.S., et al., In silico analysis of antibiotic resistance genes in the
gut microflora of individuals from diverse geographies and age-groups.
PLoS One, 2013. 8(12): p. e83823.
39. Clemente, J.C., et al., The microbiome of uncontacted Amerindians. Sci
Adv, 2015. 1(3).
40. Zhang, L., et al., Acquired antibiotic resistance: are we born with it? Appl
Environ Microbiol, 2011. 77(20): p. 7134-41.
41. Coyne, M.J., et al., Evidence of extensive DNA transfer between
bacteroidales species within the human gut. MBio, 2014. 5(3): p. e01305-
14.
42. Jakobsson, H.E., et al., Short-term antibiotic treatment has differing long-
term impacts on the human throat and gut microbiome. PLoS One, 2010.
5(3): p. e9836.
43. Perez-Cobas, A.E., et al., Differential effects of antibiotic therapy on the
structure and function of human gut microbiota. PLoS One, 2013. 8(11): p.
e80201.

32

44. Jernberg, C., et al., Long-term ecological impacts of antibiotic
administration on the human intestinal microbiota. ISME J, 2007. 1(1): p.
56-66.
45. Dethlefsen, L., et al., The pervasive effects of an antibiotic on the human
gut microbiota, as revealed by deep 16S rRNA sequencing. PLoS Biol,
2008. 6(11): p. e280.
46. Dethlefsen, L. and D.A. Relman, Incomplete recovery and individualized
responses of the human distal gut microbiota to repeated antibiotic
perturbation. Proc Natl Acad Sci U S A, 2011. 108 Suppl 1: p. 4554-61.
47. Lindgren, M., et al., Prolonged impact of a one-week course of
clindamycin on Enterococcus spp. in human normal microbiota. Scand J
Infect Dis, 2009. 41(3): p. 215-9.
48. Nyberg, S.D., et al., Long-term antimicrobial resistance in Escherichia coli
from human intestinal microbiota after administration of clindamycin.
Scand J Infect Dis, 2007. 39(6-7): p. 514-20.
49. Ghibellini, G., E.M. Leslie, and K.L. Brouwer, Methods to evaluate biliary
excretion of drugs in humans: an updated review. Mol Pharm, 2006. 3(3):
p. 198-211.
50. Hickling, D.R., T.T. Sun, and X.R. Wu, Anatomy and Physiology of the
Urinary Tract: Relation to Host Defense and Microbial Infection. Microbiol
Spectr, 2015. 3(4).

33

51. Zhang, L., et al., Antibiotic administration routes significantly influence the
levels of antibiotic resistance in gut microbiota. Antimicrob Agents
Chemother, 2013. 57(8): p. 3659-66.
52. Chantziaras, I., et al., Studying the effect of administration route and
treatment dose on the selection of enrofloxacin resistance in commensal
Escherichia coli in broilers. J Antimicrob Chemother, 2017. 72(7): p. 1991-
2001.

34

Chapter 2: Assembly of the Hospital-Based Cohort

35

2.1 ABSTRACT
Data were collected from Intensive Care Unit (ICU) patients of the Los
Angeles County + University of Southern California Medical Center (LAC+USC
MC) between 1993 and 2003 from the hospital Affinity Electronic Medical Record
(EMR) database and an ICU EMR database. Data were extracted and loaded as
Microsoft ACCESS databases, subsequently converted into Microsoft Excel
workbooks for clinical review, and imported into SAS as sas7bdat files for data
cleaning, manipulation, and assembly. The independent variable was the route of
elimination of antibiotics administered and classified using medication
administration data from the ICU EMR. The dependent variable was drug
resistance of bacteria and based on susceptibility data of organisms identified
from blood cultures. Demographic characteristics, death, and select
comorbidities were derived for all patients in the ICU. The final analytic dataset
had 11,576 patients, of which 9,541 were administered at least one antibiotic and
2,035 had no antibiotic use. A total of 10,497 patients had no eligible blood
cultures. Of those with a positive blood culture (N=1,079), 566 were non-
resistant, 398 multi-drug resistant, and 115 extensive-drug resistant.
2.2 INTRODUCTION
Data collected from all patients in the LAC-USC MC ICU and other units of
the hospital comprised of numerous datasets which were stored in two relational
databases, extracted, and combined for the purposes of analyses. Several
processes were required for the assembly of this data into analysis-ready
datasets including data cleaning, expert review, and joining (one-to-one, one-to-

36

many) raw data into analysis-ready datasets. The purpose of this document is to
provide detailed methods, assumptions, and programming specifications of the
analysis-ready datasets required for our analyses.
2.3 DATA COLLECTION
Data were collected for Intensive Care Unit (ICU) patients in LAC-USC
MC between 1993 and 2003 from a hospital Affinity Electronic Medical Record
(EMR) database that captured patient history for all patients in the hospital and
an ICU EMR database that was limited to patient history among patients in the
ICU. The Affinity EMR database generated a patient identifier and captured
detailed information for each inpatient hospital admission including admission
date and time, admitting diagnosis, admitting physician, date of birth, sex, race,
medical history, surgeries and procedures, discharge disposition, total billing
amount, discharge date, and subsequent patient readmissions. Blood culture and
microorganism susceptibility results were collected for all patients in the hospital
with a positive blood culture in the study period. The ICU EMR database
generated a new ICU patient identifier for each new admission into the ICU,
resulting in 1 to multiple ICU patient identifiers for each patient. Information
collected during the ICU stay included ICU admission date and time, sex, race,
weight, prescribed medications, ventilation, intubation, and catheter use, blood
transfusions, and laboratory parameters by organ site. Medications information
captured in the ICU database included verbatim medication name, dosing date
and time, dose, and route of administration. For catheter lines administered in the
ICU, type of catheter, fluid volume, and date of administration were recorded.

37

The data were extracted from the Affinity and ICU EMR databases and
loaded as Microsoft ACCESS databases containing nine medications datasets by
calendar year; seventy laboratory datasets by organ site (kidney function, liver
function, hematology, shock, pancreatic function, diabetic control, and
disseminated intravascular coagulation) and calendar year; seventy catheter
datasets by type of line (arterial blood gas, blood transfusion, central venous
pressure, hemodialysis, pulmonary artery pressure, total parenteral nutrition,
ventilator) and calendar year; a dataset of microorganism identified from positive
blood cultures and susceptibility data; admission and discharge data for the
inpatient hospitalization stay; and admission and discharge data for the ICU stay.
All Microsoft ACCESS databases were converted into Microsoft Excel
Workbooks and imported into SAS as sas7bdat files. Among multiple datasets
containing potentially repeat information, the SAS procedure proc compare was
used to compare the contents of the datasets.
2.4 PROCESS FLOW
Assembly of the final analysis dataset involved identification of patients in
the ICU, followed by identification of patients with antibiotic administration among
those in the ICU, and separately, classification of positive blood cultures among
patients in the ICU (Figure 1). These two datasets were combined to classify the
exposure, antibiotic elimination route, and outcome, drug resistance, among
patients in the ICU. All antibiotics contributing to the exposure classification were
identified. Select comorbidities were assessed for patients in the ICU.

38

Demographic and patient, exposure, outcome, and comorbidity variables were
combined to form a single final analytic dataset.
2.5 PATIENTS IN THE ICU
Data structure: one record per patient per hospital and ICU admission
Methods
Contiguous or overlapping inpatient hospitalization stays were combined
into a single inpatient stay. All hospital and ICU admission records were
combined for patients with both inpatient hospitalization stays and ICU stay and
patients with ICU stays but no inpatient hospitalization stay. Demographic and
patient characteristics for all patients in the ICU.
Programming
Each patient admitted into the hospital had an 8-digit unique hospital
patient identifier generated, which is used for all hospital admissions at LAC-USC
MC. Any subsequent readmissions into the hospital were linked to the unique
hospital patient identifier in the Affinity database. Admission into the ICU
generated a separate 4-digit ICU patient identifier that changed for each
subsequent patient readmission into the ICU, so that a person admitted to the
hospital had n ICU identifiers for each of the n times the patient was admitted into
the ICU, regardless of whether this occurred during a single inpatient hospital
stay.
Inpatient hospitalization stays were bridged and contained the earliest
admission date and latest discharge date (Figure 2) according to the following
algorithm:

39

 For multiple records with admission on the same day, the record
corresponding to the latest discharge date was selected.
 If an inpatient stay record contained both admission and discharge dates
while a subsequent inpatient stay record contained only the admission
date with a missing discharge date, the inpatient stay record containing
complete admission and discharge dates was kept.
 If a patient had multiple inpatient stays in which the admission dates
remained the same but discharge dates differed, the maximum discharge
date was taken.
The ICU patient experience was linked to the hospital patient experience
using hospital and ICU patient identifiers and admission dates as keys. Among
patients with both an inpatient hospitalization and ICU stay, records were linked
based on the occurrence of ICU admission dates between a hospital admission
and subsequent hospital readmission. If a patient had no subsequent
readmission, linkage was done based on the presence of an ICU admission on or
after the hospital admission. Examination of the raw data showed that the ICU
discharge dates did not always precede (defined as being prior to or the same
day as) the hospital discharge date. Therefore, due to this potential discordance,
discharge dates were not used to link hospital and ICU patient identifiers. Among
patients with an ICU stay but no inpatient hospitalization stay, hospital admission
and discharge dates were set to missing. In some instances, patients with both
hospital and ICU admissions had hospital admission records which contained
only the admission and discharge dates and no other patient or hospital

40

characteristics. Such records were kept in the dataset for patients in the ICU but
were not utilized for the derivation of exposure, outcome, or downstream
analyses.
Demographic and hospital characteristics from both inpatient
hospitalization stays and ICU stays were kept. Length of stay (LOS) was defined
as [(discharge date – admission date)+1] for hospital and ICU stays. Cumulative
hospital and ICU LOS was defined as the sum of the hospital and ICU LOS over
the entire patient experience in the study period. Admitting diagnoses into the
hospital were captured in the data as ICD-9-CM diagnosis codes. In order to link
these codes to diagnoses, a codelist of diagnosis codes and corresponding
diagnoses were captured for all ICD-9-CM codes according to diagnosis groups
in an Excel spreadsheet, as specified by the Centers for Medicare and Medicaid
Services (CMS) [1]. The codelist was merged to admitting diagnosis codes in the
ICU cohort using the ICD-9-CM code as the key.
Occurrence of death was identified separately for patients during the
hospitalization stay and during the ICU stay. If the route of discharge from the
hospital indicated that the patient was deceased, then the patient was considered
to have died during the hospital admission. If discharge from ICU is to the
morgue then patient was considered to have died during the ICU stay. If a patient
died during the ICU stay, the date of death was assumed to be the date of
discharge from the ICU; otherwise, if a patient died during the inpatient hospital
stay but not during the ICU stay, the date of death was assumed to be the
discharge date from the hospital. If a patient’s record indicated that the patient

41

died during the ICU stay and the hospital stay, and the discharge dates were
different, then the earliest of the hospital and ICU dates was taken as the date of
death. For patients whose discharge data did not indicate death, the patients
were classified as not being deceased while under observation, and the date of
death was set as null.
The final dataset had 17,017 records and contained hospital identifier, ICU
identifier, admission year, age, sex, race, weight, hospital admission and
discharge dates, ICU admission and discharge dates, hospital LOS, ICU LOS,
admitting diagnoses, discharge status, death during the hospitalization stay,
death during the ICU stay, and death date,.
2.6 PATIENTS WITH ANTIBIOTIC USE
Data structure: one record per patient per hospital and ICU admission and
antibiotic administration
Methods
Antibiotics administration records were extracted from yearly medications
datasets which ranged from 1995-2003. The route of elimination was classified
for antibiotics as “predominately through the kidneys” or “predominantly through
the liver” based on whether dose adjustment of the antibiotic was required for
patients with compromised renal function. If dose adjustment of the antibiotic was
required, the antibiotic was classified as being eliminated predominately through
the kidneys; otherwise, if dose adjustment of the antibiotic was not required, the
antibiotic was classified as being eliminated predominantly through the liver.
Dose adjustment was identified from antibiotic packet insert information [2].

42

Antibiotic administration records and route of elimination were linked back to
patients in the ICU cohort.
Programming
Annual medications datasets containing ICU patient identifier, verbatim
medication name, dose, and medication administration start and end date were
combined into a single medications dataset (N=426,806 records) using ICU
patient identifier as the key. From this dataset, a listing of unique verbatim
medication terms was extracted into an Excel spreadsheet. Verbatim names
were mapped to branded and generic terms and received by the study clinician
and study pharmacologist. A review of the generic and branded medication
names was done to identify antimicrobials (yes/no) and antibiotics (yes/no), from
which sixty-eight unique antibiotics were identified. Route of antibiotic elimination
was classified as either “through the kidneys” or “through the liver” for each
antibiotic. The antibiotics and corresponding routes of elimination were reviewed
by a LAC-USC MC clinician and separately by a LAC-USC MC pharmacologist.
Once review of the spreadsheet containing verbatim, generic, and branded
medications, antimicrobial flay (yes/no), antibiotic flag (yes/no), and route of
elimination for antibiotics was finalized, the spreadsheet was imported into SAS,
and the variables were linked to the set of all medications using the generic
medication term as the key. A record-level antibiotic flag (yes/no) was derived to
denote the presence of an antibiotic record. Antimicrobials other than antibiotics
were identified and reviewed, but were not kept for the purposes of this analysis.

43

The medications dataset was then restricted to all antibiotic records by
selecting on the antibiotics flag (N=53,280 records). The ICU cohort (one record
per patient and ICU admission) was combined with the antibiotics dataset (one
record per patient per antibiotic administration) in order to characterize antibiotic
usage among ICU patients. For patients with antibiotic use, the antibiotic
administration start and stop date and elimination route for each antibiotic record
were kept. For ICU patients with no antibiotic use, the dataset was padded to
contain a single record with null values for administration start date, stop date,
and elimination route. The datasets were joined using the following algorithm:
 Patients in the ICU with antibiotic administration: identified as the
presence of an ICU identifier in both the ICU cohort and the antibiotics
dataset, operationalized by an inner join of the antibiotics dataset to the
cohort with the ICU patient identifier as the key.
 Patients in the ICU with no antibiotic administration: identified as the
presence of an ICU identifier in the ICU cohort with an absence of the ICU
identifier in the antibiotic administration dataset. This was operationalized
by left joining the antibiotics dataset to the cohort using the ICU patient
identifier as the key, where the key identifier was missing from the
antibiotics usage dataset but not missing in the cohort dataset.
 Patients in the ICU cohort with no ICU identifier (these patients had only
hospital identifier) is a group for which antibiotic usage cannot be
assessed because of patients with missing ICU enrollment information.

44

These patients were categorized as having missing antibiotic exposure
information.
A patient-level antibiotics flag was created to indicate the presence
(yes/no) of at least one antibiotic administration for each patient. The final
antibiotics dataset had 48,568 records and contained hospital identifier, ICU
identifier, hospital and ICU admission start and end dates, demographic and
patient characteristic variables identified above, generic name of antibiotic,
antibiotic administration start and end dates, route of elimination, a patient level
flag indicating no antibiotic information, and a patient level flag indicating
presence of any antibiotics.
2.7 PATIENTS WITH POSITIVE BLOOD CULTURES
Data structure: one record per patient per hospital and ICU admission and blood
culture accession number and organism
Methods
Positive blood cultures were collected for all patients with an inpatient
hospitalization stay exhibiting signs and symptoms of infection during the study
period. Data contained information of results of the blood culture and results of
susceptibility testing against a panel of antimicrobial agents. All microorganism
information other than susceptibility results was extracted from the blood culture
dataset and linked back to patients in the ICU. Patients were classified for the
presence of a positive blood culture.
Programming

45

Microbiology and susceptibility results were collected for hospitalized
patients with positive blood cultures between 1994 and 2003. The raw dataset
contained information on blood culture accession date, blood culture accession
number, genus and species of organism cultured, gram stain classification of
organism, and organism susceptibility profile. A dataset containing only blood
culture information among all hospitalized patients was derived by extracting the
unique hospital patient identifier, accession date, accession number, and
organisms. All isolates were extracted and output as an Excel spreadsheet for
review. Non-bacterial isolates (virus, fungi), ill-defined microorganisms, or
microorganisms resulting from probable contamination were marked for
exclusion and reviewed independently by the study clinician and pharmacologist.
The final spreadsheet containing isolate overrides was imported into SAS. Only
isolates not marked for exclusion were kept in the data. Gram stain classification
was reviewed and incorrect gram stain classification in the raw data was re-
categorized where necessary. Accession dates and numbers were reviewed as
part of data cleaning. Records in which organisms had the same accession
number with multiple accession dates were identified and examined; however,
these were determined to be an artifact of the data collection method and not
data error, and thus no cleaning was performed.
We limited this microorganism dataset to patients in the ICU. Linkage of
the culture information (one record per patient per isolate record) to the patients
in the ICU (one record per patient and ICU admission) was done using hospital
patient identifier as the key. For each patient, we checked whether a blood draw

46

tested positive for an organism during an inpatient hospitalization stay. This was
operationalized using blood culture accession dates and hospital admission
dates: only accession dates that were between a patient’s hospital admission
and subsequent hospital readmission dates, or, if a patient had no subsequent
readmission, between the hospital admission and discharge dates, were kept.
Among blood culture records which did not merge to patients in the ICU cohort
based on this criteria, records were linked on the hospital patient identifier only.
For patients with a positive blood cultures, the accession number,
accession date, organism, and gram stain classification for each isolate were
kept. For ICU patients with no positive blood cultures, the blood culture dataset
was padded to contain each patient and ICU admission record, with
corresponding null values for accession number, accession number, organism,
and gram stain classification. Blood cultures were numbered in ascending order
based on accession date. If a patient had no positive blood cultures and had only
records for ICU admissions, a row number was assigned to each admission in
ascending order. A patient-level blood culture flag (yes/no) was created to
indicate the presence of a positive blood culture, and the total number of positive
blood cultures for each patient was derived. The final organism dataset had
19,430 records and contained hospital identifier, ICU identifier, hospital and ICU
admission start and end dates, demographic and patient characteristic variables
previously identified, blood culture accession date, blood culture accession
number, genus and species of the organism, gram stain classification, row

47

number, total number of positive blood cultures, and a patient-level flag indicating
presence of any positive blood cultures.
2.8 PATIENTS IN ICU WITH ANTIBIOTIC AND RESISTANCE
CLASSIFICATIONS
Methods
The final analytic dataset was limited to patients with complete
demographic, microbiology and medication data, ranging from October 1, 1995
until December 31, 2002. Patients with data only prior to October 1, 1995 were
excluded from the dataset. Type of ICU (medical, surgical, neurosurgical) was
not provided for patients in the ICU in 2003; therefore, the study period was
truncated to 2002. A HA-BSI was defined as a bloodstream infection occurring at
least 3 days after admission to the hospital (Figure 3). Organisms in the blood
culture dataset were categorized based on whether the hospital-acquired criteria
and latency period criteria were met. A latency period of 2 days was required
between antibiotic exposure and resistance development of organisms identified
from blood cultures. Details of this will be described in the sections to follow.
Among all positive blood cultures, only qualifying organisms, defined as those
meeting the hospital-acquired and latency period criteria, contributed to the
resistance score. If a patient had no organism that met both the hospital-acquired
criteria and the latency period criteria, then the patient was considered as not
having any qualifying organisms; otherwise the patient had at least one qualifying
organism.

48

2.8.1 Patients in ICU with Antibiotic Classifications
Data structure: one record per patient and blood culture accession number and
organism
Methods
The route of antibiotic elimination was classified as ‘no antibiotic
exposure’, ‘antibiotics only eliminated predominantly through the kidneys’, or ‘any
antibiotics eliminated at least partially through the liver’. Among patients with
blood cultures, the route of antibiotic elimination was summarized for each blood
culture. Among patients with no blood culture, the route of elimination was
summarized once for each patient. All antibiotic administrations that started at
least two days prior to blood culture sampling were used to determine the
elimination route corresponding to each qualifying blood culture. We classified
the elimination route for patients receiving antibiotics only eliminated through the
kidneys up to the time of their blood culture as ‘through the kidneys’. We
classified the elimination route for patients receiving any antibiotics eliminated at
least partially through the liver prior as ‘through the liver’ for the associated blood
culture. Elimination through the liver was assumed to be cumulative and was
carried forward, so that patients receiving any antibiotics eliminated through the
liver prior to a qualifying blood culture were classified as having non-renal
elimination for the qualifying and all subsequent blood cultures.
Among patients with no blood cultures taken or no isolates grown from
blood cultures, antibiotic elimination was defined using all antibiotics during their
ICU experience. Patients with no antibiotic use during their ICU stay had no

49

antibiotics contributing to the exposure definition and thus were classified as
having no antibiotic exposure. Among patients with antibiotic use and positive
blood cultures, all antibiotic use up to each qualifying blood culture were
considered. If a patient with antibiotic use and positive blood culture had no
qualifying blood culture, all antibiotics in the duration of the ICU stay were
considered and the patient was classified as having no organism.
Programming
To identify route of antibiotic elimination and resistance classification
among patients in the ICU, patients in the antibiotics dataset were combined with
patients in the microbiological dataset for each of the following groups: patients
with no antibiotic administration and no positive blood cultures; no antibiotic
administration and positive blood cultures; antibiotic administration and no
positive blood cultures; and both antibiotic administration and positive blood
cultures.
Patients with either antibiotic administration or positive blood culture, or
neither, were joined from the antibiotics and blood culture datasets using the
hospital patient identifier and hospital admission date as keys. The resulting
dataset was one record per patient per admission stay and blood culture and
antibiotic administration. Because of the fuzzy merge between antibiotic records
and blood culture records, we examined each record for the presence of
antibiotics in order to classify the exposure. If antibiotic administration was
present for the record, then the record contributed to the derivation of the overall
antibiotic exposure.

50

 No antibiotic and no blood culture: If the record had no accession date and
no drug start date, then the record was not flagged for inclusion because
there was no antibiotic that contributed to the cumulative antibiotic
exposure.
 Antibiotic administration and no blood culture: If the record had no
accession date and a drug start date, then this record was flagged for
inclusion in the cumulative antibiotic exposure.
 No antibiotic and no blood culture: If the record had an accession date and
no drug start date, then this record was not flagged for inclusion because
of the lack of antibiotic that could contribute to the cumulative antibiotic
exposure
The cumulative route of antibiotic elimination was then classified as follows:
 If patients had no records flagged for inclusion, indicating no antibiotic
administrations, the cumulative antibiotic exposure was categorized as “no
antibiotic exposure”.
o If patients had no blood culture, then the exposure was classified
for each ICU admission.
o If patients had a positive blood culture, the exposure was
categorized for each blood culture record.
 If patients had at least one record flagged for inclusion, indication
antibiotic administration, then

51

o If any of the antibiotics were eliminated at least partially through the
liver up to the ICU discharge date, then the cumulative antibiotic
exposure was classified as “through the liver”.
o Otherwise, if all of the antibiotics were eliminated predominantly
through the kidneys up to the ICU discharge date, then the
cumulative antibiotic exposure was classified as “through the
kidneys”.
Patients with both antibiotic administrations and positive blood cultures
were treated as a separate group for which additional processing was required,
and were combined using a Cartesian join with only the hospital patient identifier
as the key. The Cartesian join allowed for the identification of antibiotics
contributing cumulatively for qualifying organisms using all antibiotic-organism
records.
Since only qualifying blood cultures were considered for the antibiotic
classification among patients with antibiotic administration and positive blood
culture, we determined whether each organism met the HA-BSI and latency
period criteria prior to examining each antibiotic-organism pairing. Organisms
were classified as being hospital-acquired (yes/no) if the associated blood culture
accession date was at least 3 days after the inpatient admission date. If the
inpatient admission date was missing, the ICU admission date was used to
determine whether organisms met this criteria. Blood cultures were considered to
meet the latency criteria if the blood culture accession date occurred on or after
the antibiotic administration start date +2 days. Specifically, we flagged records

52

which did not meet the latency criteria with respect to antibiotic exposure if the
accession date occurred on or preceding the antibiotic start date +1 day.
After identifying qualifying organisms, we flagged all antibiotic
administrations that preceded a blood culture accession date by at least two days
in order to identify which antibiotics contributed to the exposure derivation for a
specific organism. Antibiotic administration records whose start and end dates
occurred before a positive blood culture, or antibiotic administration record start
and end dates during which a positive blood culture occurred were considered;
antibiotics whose administration occurred after the accession date of the
organism were not considered.
Because this generated a fuzzy merge resulting in all possible antibiotic-
organism pairing, all records were examined for the specified criteria. Among
patients with a qualifying blood culture and for each antibiotic-organism record,
antibiotics were flagged (yes/no) to indicate whether it contributed towards the
cumulative antibiotic exposure, using the following algorithm:
 If a person had no organisms that met the HA-BSI criteria
o Then, for each positive blood culture, all antibiotic administrations
were flagged for inclusion
 If no antibiotic administration-blood culture combinations meet the latency
criteria because all of a person’s antibiotic administrations were in a -1day
or greater window relative to the positive blood culture, then all antibiotic
administrations up to the discharge date were flagged for inclusion

53

 If the record had a qualifying accession date and a drug start date, then
the record was flagged for inclusion in the cumulative antibiotic exposure if
the accession date occurred on or after two days after the drug start date.
o If the qualifying organism preceded antibiotic administration
because the accession date occurred before the drug start date,
then the antibiotic was not flagged for inclusion
o If the accession date occurs on the drug start date or on the day
after the drug start date, then the antibiotic is not flagged for
inclusion
o If the accession date occurred at least two days after drug start
date and prior to or including the drug end date, then the antibiotic
record was flagged for inclusion
o If the accession date occurred at least two days after drug start
date and after the drug end date, then the antibiotic record was
flagged for inclusion
For each patient with antibiotic use and qualifying blood culture, the
cumulative antibiotic exposure was classified based on the following algorithm,
using only antibiotic records which were flagged for inclusion:
 If any antibiotics up to the qualifying blood culture were eliminated at least
partially through the liver, then the cumulative antibiotic exposure was
classified as “through the liver” for the blood culture .

54

 If all of the antibiotics up to the qualifying blood culture were eliminated
predominantly through the kidneys, then the cumulative antibiotic
exposure was classified as “through the kidneys” for the blood culture.
If a patient had no qualifying blood culture, then all antibiotic administrations
contributed to the cumulative antibiotic exposure but no positive blood cultures
were considered for resistance.
The dataset of contributing antibiotic administrations among patients with
antibiotic administration, positive blood cultures, or neither was combined with
the dataset of contributing antibiotic administrations among patients with
antibiotic administration and positive blood cultures. The resulting dataset had
13,696 records and contained hospital identifier, ICU identifier, demographic and
patient characteristic variables previously identified, earliest administration start
date and latest end date of contributing antibiotics, final exposure variable that
summarized route of antibiotic elimination, a patient level flag indicating no
antibiotic information, patient level flag indicating presence of any antibiotics,
blood culture accession date, blood culture accession number, genus and
species of the organism, gram stain classification, a flag for each organism to
indicate whether it was a qualifying organism, and a patient-level flag indicating
presence of any positive blood cultures. Only hospital admissions during which
antibiotic administrations or positive blood cultures were present, and
corresponding discharge date, were captured in this dataset.

55

2.8.2 Patients in the ICU with Resistance Score Classification
Data structure: one record per patient and blood culture accession number and
organism
Methods
Organisms were tested against a panel of antimicrobial agents.
Resistance scores were derived for each patient and qualifying organism using
methods specified by Magiorakos, et al. [3]. First, microorganisms were
categorized as one of five epidemiologically significant bacterial groups:
Staphylococcus aureus, Enterococcus species, Enterobacteriaceae family,
Pseudomonas aeruginosa, or Acinetobacter species. Gram negative bacteria not
categorized as Enterobacteriaceae family, Pseudomonas aeruginosa, or
Acinetobacter species were imputed as one of the three groups above based on
the drug resistance profile; otherwise they were categorized as a “general gram
negative” bacteria. Similarly, gram positive isolates identified from blood cultures
were categorized as Staphylococcus aureus or Enterococcus species, imputed
as Staphylococcus aureus or Enterococcus species based on their resistance
profile, or classified as “general gram positive” if neither. The final organism
mapping underwent review by the study pharmacologist.
Multi-drug, extensive-drug, and pan-drug resistance (MDR, XDR, and
PDR) resistance was derived for each bacterial group. Multi-drug resistance was
defined as resistance to at least one antimicrobial agent in at least three
antimicrobial classes; extensive-drug was defined as resistance to at least one
antimicrobial agent in all but two or less antimicrobial classes; and pan-drug

56

resistance was defined as resistance to all antimicrobial agents in all classes. For
each category of bacterial group, only select antimicrobial categories and agents
were used to define MDR, XDR, and PDR. Specifically, only susceptibility data
for antimicrobial classes with activity against the specified bacterial group
contributed to the resistance definition for each bacterial group. Bacteria with
intrinsic resistance to an antibiotic or antimicrobial class did not contribute to the
resistance derivation. In addition, organisms that did not come from a qualifying
blood culture did not contribute to the resistance derivation. For patients with a
polymicrobial culture, the most resistant score and corresponding organism was
used for all analysis. Among patients with multiple blood cultures over time, the
organism and corresponding resistance score from the first qualifying blood
culture was selected and used for all analyses.
Programming
Data cleaning and manipulation of the susceptibility dataset was
performed. Patient identifiers from the susceptibility dataset were padded with
additional zeros in order to be comparable in character length to the patient
identifiers found in all other datasets. Resistance scores were provided as
character variables in the raw data; a numeric version of the variable was created
based on the following ordinal categorization: resistance-1, intermediate-2; and
susceptible-3. For distinct records of blood culture accession date, accession
number, and organism, the presence of duplicate and multiple susceptibility
results for an antimicrobial agent was assessed. Among records with multiple

57

susceptibility results, the most resistance result was kept. Among records with
duplicate resistance scores for the same agent, a single record was kept.
The dataset of susceptibility results was combined with two dataset: one
containing the route of antibiotic elimination classification among all patients in
the ICU and the other containing positive blood cultures among all patients in the
ICU. The three datasets were merged using the hospital patient identifier, blood
culture accession number, blood culture accession date, and cultured organism
as keys. Due to the potential of organisms being tested against antimicrobial
agents against which they had no activity, as determined by on gram stain
classification, the list of agents used for gram positive organisms and the list of
agents used for gram negative organisms was reviewed. Unique agents were
provided in an Excel spreadsheet for gram positive and gram negative
organisms. Based on review by the study pharmacologist, the following data
cleaning of the susceptibility results was done:
 For gram negative organisms, all susceptibility result records for the
antimicrobial agent clindamycin were removed, because clindamycin is
administered for gram positive organisms.
 For gram positive organisms, all susceptibility result records for the
antimicrobial agent amikacin were removed, because this antibiotic has
activity against gram negative organisms only.
 Finally, all susceptibility result records for testing was done against
extended spectrum beta (ß) lactamase (ESBL), a bacterial enzyme and
not an antimicrobial agent, were removed.

58

After data cleaning, organism resistance scores were derived as
described. First, all organisms and their gram stain classifications were output
into an Excel spreadsheet. For each of the 113 unique organisms identified from
all positive blood cultures, organisms were mapped to one of the bacterial
groups: Acinetobacter species, Enterobacteriaceae family, Enterococcus
species, Pseudomonas aeruginosa, or Staphylococcus aureus based on gram
stain classification and organism activity [3]. Organisms that were not suited to
be mapped to these categories were classified as “gram positive” or “gram
negative”. The mapped bacterial group was reviewed by the study
pharmacologist, and the final mapping was merged back onto the dataset
containing organism and resistance information using the organism name and
gram stain classification as key identifiers. The result was an additional column
which contained the mapped bacterial group corresponding to each organism in
the data.
Next, to determine which antimicrobial agents from the susceptibility
results contributed to the resistance definitions for each of the seven bacterial
groups, a listing of unique antimicrobial agents was exported as an Excel
spreadsheet. Each of the 56 agents were classified into antimicrobial classes
separately for each of the bacterial groups, based on activity against each group.
This was operationalized in the Excel spreadsheet by creating a column for each
of the bacterial groups. The seven columns contained the mapped antimicrobial
class based on information provided in Margioakos, et al. [3]. For each bacterial
group, antimicrobial agents which had no activity to the specific group did not

59

contribute a resistance score against that organism and were marked for
exclusion in the spreadsheet [3]. The spreadsheet contained the antimicrobial
agent-to-class mapping among antimicrobial agents contributing to the resistance
for each bacterial group, and was reviewed by the study pharmacologist. The
final spreadsheet was imported into SAS and combined with the above dataset
using antimicrobial agent as the key. The result was an additional column that
contained antimicrobial classes against which the bacterial group corresponding
to an organism had activity. Records of antibiotic agents that had no activity
against the bacterial group were deleted from the dataset. Additionally, a single
isolate identified as anaerobic cocci was deleted because of the inability to
accurately determine a resistance profile for an unspecific organism. Finally,
agents tested against each bacterial group were reviewed for intrinsic resistance
against the group; if the bacterial group had intrinsic resistance against the agent
or class tested, the corresponding record in the dataset was marked for
exclusion. All records marked for exclusion were deleted. Antimicrobial classes
were classified as “resistance” if the susceptibility result was “resistant” or
“intermediate”; otherwise, antimicrobial classes were classified as “susceptible”.
For each organism, the total number of contributing antimicrobial classes was for
the determination of the resistance score, and thus stored as a separate variable.
For each patient, organism, and antimicrobial class, the most resistant
susceptibility result was taken. MDR, XDR, and PDR were derived for each
organism using the following algorithm:

60

 If at least 3 antimicrobial classes were present in the susceptibility panel
for the organism:
o If all antibiotic classes were resistant, then the organism was
classified as PDR.
o Otherwise, if an organism was resistant to at least 3 antimicrobial
classes but not resistant to 2 or less antimicrobial classes, then the
organism was classified as XDR.
o Otherwise, if an organism was resistant to at least 3 resistance
antimicrobial classes, then the organism was classified as MDR.
 If less than 3 antimicrobial classes were present in a susceptibility panel,
then the resistance score for the organism was classified as missing.
The final classification of organism resistance was merged onto the dataset of
ICU patients containing classification of the antibiotic elimination route using the
accession date, accession number, and the organism name as keys. The
following algorithm was used to create the final resistance score:
 Among patients with no antibiotic use and no blood cultures, the single
patient-level record was classified as “no organism”.
 Among patients with antibiotic use and no blood cultures, the single
patient-level record was classified as “no organism”.
 Among patients with blood cultures, regardless of antibiotic use:
o If a patient had at least one qualifying blood culture,
 The resistance score for the organism was based on MDR,
XDR, and PDR calculations derived above.

61

 Otherwise, if the organism was not MDR, XDR, or PDR, the
organism resistance score was classified as “no resistance”.
o If a patient had no qualifying organisms, the resistance score was
classified as “no organism” for each of the organisms; additionally,
the organisms were renamed as “no organism”.
Although patients had potentially multiple blood cultures over time, only the first
qualifying organism was of interest for our current analyses. Records for the first
qualifying organism were identified using the following algorithm:
 Among patients with no antibiotic use and no blood cultures, the single
patient-level record was flagged for inclusion.
 Among patients with antibiotic use and no blood cultures, the single
patient-level record was flagged for inclusion.
 Among patients with blood cultures, regardless of antibiotic use :
o If a patient had at least one qualifying blood culture, the record
corresponding to first qualifying organism was flagged for inclusion.
o If a patient had no qualifying blood culture, then the first record was
flagged for inclusion since the exposure did not vary and the
outcome was reclassified as “no organism” for all records.
Finally, the index date was defined as the earliest or non-missing instance
of the antibiotic administration start date or blood culture accession date, among
patients with antibiotics, positive blood cultures, or both. For patients with neither,
the index date was defined as the hospital admission date. The resulting dataset
had 13,538 records and contained hospital identifier, ICU identifier, demographic

62

and patient characteristic variables previously identified, antibiotic variables
previously identified, blood culture variables previously identified, resistance
score, index date, and the first qualifying organism flag.
2.9 PATIENTS IN ICU WITH COMORBIDITY CLASSIFICATIONS
Data structure: one record per patient and blood culture accession number and
organism
Methods
All diagnoses and procedures during inpatient hospital stays were
recording using ICD-9-CM diagnosis codes, ICD-9-CM procedure codes, and
Current Procedural Terminology (CPT) codes. Diagnoses codes were extracted,
linked to all patients in the ICU, and further linked to a comorbidities codelist to
determine the presence of comorbidities of interest. The presence of
comorbidities was assessed in a one-year lookback period from the index date.
Patients with multiple instances of the same type of comorbidities in the lookback
period were counted only once for the presence of that comorbidity.
Programming
Records were pulled from data containing up to 20 ICD-9-CM diagnosis
codes, up to 20 ICD-9-CM procedure codes, and up to 20 CPT codes
corresponding to all diagnoses and procedure that occurred during a patient’s
hospital experience. Distinct records grouped by hospital identifier, hospital
admission and discharge date, admitting diagnosis, and total billed among of the
hospital stay were transposed using these same variables as keys and keeping
only ICD-9-CM diagnosis codes to generate a long dataset containing one record

63

per patient per inpatient hospital stay and diagnosis code. Empty cells that were
generated during the transposition of the data, indicating fewer than 20
diagnoses for an inpatient stay, were removed.
The dataset of patients in the ICU with exposure and outcome
classifications was combined with the dataset of patients with ICD-9-CM
diagnosis codes using hospital identifier as the key. This dataset was then joined
with the comorbidities codelist, which was imported from an Excel spreadsheet,
using ICD-9-CM diagnosis codes as the key. We limited the data to records
where the comorbidity diagnoses occurred after the index date -1 year but prior
to the index date (inclusive). The resulting dataset had one record per person per
inpatient hospitalization and comorbidity, for up to multiple comorbidities within
each hospitalization stay. A patient could have multiple diagnosis codes for the
same comorbidity; therefore distinct records for each patient, index date,
comorbidity category, and comorbidity were taken. The resulting dataset had
19,366 records.
2.10 PATIENTS IN THE FINAL ANALYTIC DATASET
Data structure: one record per patient and blood culture accession number and
organism
Methods
Demographic and hospital characteristics, route of elimination, resistance
score, and comorbidities were combined for all patients in the ICU.
Programming

64

Comorbidities were appended to the dataset of ICU patients containing
exposure and outcome classification through a series of left-joins, for the
generation of binary variables indicating the presence or absence of the specified
comorbidity in the one-year lookback period relative to each index date. For each
hospital identifier and index date, we generated flags for the presence of the
following comorbidities with each comorbidity category:
 Bone: Fracture, Osteoporotic fracture, Osteoporosis
 Cancer: Cancer, Non-melanoma skin cancer
 Cardiovascular: Angina, stable, Angina, unstable, Heart failure,
Hyperlipidemia, Hypertension, Myocardial infarction, Peripheral vascular
disease, Stroke, Transient ischemic attack, Ventricular arrhythmia
 Central Nervous System (CNS): Alzheimer’s, Dementia, Migraine,
Parkinson’s disease
 Connective Tissue Disorder: Psoriatic arthritis, Psoriasis, Rheumatoid
arthritis, Systemic lupus erythematosus
 Diabetes mellitus (Type 2)
 Gastro-intestinal: Crohn’s disease, GI Bleed, Irritable Bowel Syndrome,
Ulcerative colitis
 Hepatic: Chronic liver disease, Cirrhosis
 Pulmonary: Asthma, Chronic obstructive pulmonary disease (CODP)
 Renal: Chronic kidney disease (CKD) without dialysis, Chronic kidney
disease (CKD) with dialysis

65

The resulting final analytic dataset had 13,538 records and contained
hospital identifier, ICU identifier, demographic and patient characteristic variables
previously identified, earliest administration start date and latest end date of
contributing antibiotics, final exposure variable that summarized route of
antibiotic elimination, a patient level flag indicating no antibiotic information,
patient level flag indicating presence of any antibiotics, blood culture accession
date, blood culture accession number, genus and species of the organism, gram
stain classification, resistance score, index date, first qualifying organism flag,
patient-level flag indicating presence of any positive blood cultures, and flags for
each of the comorbidity categories presented above.
2.11 ANTIBIOTICS AND RESISTANCE PATTERNS AMONG PATIENTS IN THE
FINAL ANALYTIC COHORT
Methods
Among all patients in the ICU with the final exposure classification and
resistance score, antibiotics that contributed to the cumulative route of antibiotic
elimination were identified. Separately, the contributing antimicrobial
susceptibility results of qualifying organisms of patients in the ICU were
identified.
Programming
The two interim datasets, previously derived, which identified whether
antibiotic administration records contributed to the cumulative antibiotic score,
were combined. The dataset was limited to antibiotic records that did contribute
to the exposure classification, for total of 60,757 records. Select variables

66

(exposure classification, resistance score, and the first qualifying organism flag)
from the final analytic dataset were merged onto the set of all contributing
antibiotics; the final dataset contained hospital identifier, ICU identifier, hospital
admission and discharge date, type of ICU, admitting diagnosis, generic name of
antibiotic, generic antibiotic start and end date, generic antibiotic route of
elimination, earliest administration start date and latest end date of contributing
antibiotics, final exposure variable, patient level flag indicating presence of any
antibiotics, blood culture accession date, blood culture accession number, genus
and species of the organism, gram stain classification, resistance score, and a
patient-level flag indicating presence of any positive blood cultures.
Data cleaning and formatting of the patient identifier was done on the
susceptibility dataset was previously described. To identify the antimicrobial
susceptibility results (susceptible or resistant) of qualifying organisms of patients
in the cohort, all patients in the final analytic dataset with a qualifying organism
were combined with the entire susceptibility dataset using the hospital identifier,
organism, accession number, and accession date. The resulting dataset had
11,426 records and contained hospital identifier, hospital admission and
discharge date, type of ICU, admitting diagnosis, comorbidities, final exposure
variable, patient level flag indicating presence of any antibiotics, blood culture
accession date, blood culture accession number, genus and species of the
organism, gram stain classification, resistance score, patient-level flag indicating
presence of any positive blood cultures, antimicrobial agent against which the
organism was tested, and the susceptibility result (susceptible or resistant).

67

2.12 FIGURES
Figure 2.1. Process Flow of Analytic Datasets

68

Figure 2.2. Bridging of hospitalization stays for contiguous and overlapping
records

69

Figure 3.3. Minimum hospital stay and latency period required for qualifying
organisms

70

2.13 REFERENCES
1. Centers for Medicare and Medicaid Services. ICD-9-CM Diagnosis and
Procedure Codes: Abbreviated and Full Code Titles. 2014 [cited 2017
10Mar]; Available from:
https://www.cms.gov/Medicare/Coding/ICD9ProviderDiagnosticCodes/cod
es.html
2. Truven Health Analytics, Micromedex.
3. Magiorakos, A.P., et al., Multidrug-resistant, extensively drug-resistant and
pandrug-resistant bacteria: an international expert proposal for interim
standard definitions for acquired resistance. Clin Microbiol Infect, 2012.
18(3): p. 268-81.

71

Chapter 3: Patterns of antibiotic resistance in organisms cultured from HA-
BSIs in the ICU: a hospital-based cohort study

72

3.1 ABSTRACT
Previous epidemiologic studies have not directly addressed impact of
route of antibiotic elimination on subsequent drug resistance. We address this
question in a retrospective cohort study of 11,576 patients admitted to the
Intensive Care Unit of the Los Angeles County + University of Southern California
Medical Center from 1995 until 2002. We classified patients according to route of
elimination of antibiotic administered (none, liver, kidneys), assessed drug
resistance (no resistance, multi-drug resistance, extensive-drug resistance) of
organisms identified in subsequent hospital-acquired bloodstream infections (HA-
BSI), and estimated both associations between route of antibiotic elimination and
resistance and proportion of drug resistant HA-BSI organisms attributable to
each antibiotic type. Compared to patients administered no antibiotic, odds of
multi-drug resistance were 1.82(95% CI: 1.11, 2.97, P=0.0169)- versus 2.99(95%
CI: 1.84, 4.85, P<0.0001)-times higher in those administered antibiotics
eliminated through the kidneys and liver respectively, and odds of extensive-drug
resistance were 3.46(95% CI: 0.82, 14.68, P=0.0923)- versus 7.60(95% CI: 1.84,
31.41, P=0.0051)-times higher in these groups. Moreover, estimated resistance
attributable to antibiotics was 41.7% versus 72.8% among patients who received
antibiotics eliminated through the kidneys and liver, respectively. These results
accord with the possibility that risk of acquiring drug-resistant HA-BSI is higher
after treatment with antibiotics eliminated through the liver.

73

3.2 INTRODUCTION
Bloodstream infections (BSI) are a major cause of morbidity and mortality
in the United States [1]. It is estimated that half of all hospital associated BSIs
(HA-BSIs) occur in patients in the Intensive Care Unit (ICU) setting, and of these,
26-48% result in death [2]. It is becoming increasing difficult to treat HA-BSIs
because organisms acquired in the hospital setting exhibit antibiotic resistance
[3, 4]. Several organisms are multi-drug resistant and pan-drug resistant against
available antibiotics [5], and the pool of antibiotics with which to treat HA-BSIs is
becoming increasingly small [3, 5]. The resulting public health and economic
burden is significant, evidenced by increased rates of all-cause mortality,
infection related mortality, ICU length of stay (LOS), and medical costs
associated with treatment of patients with resistant infections [6].
Recent advances in microbiology have identified antibiotic exposure as an
important modulator of the immune function of the gastrointestinal tract, whose
lumen is a milieu of commensal and pathogenic bacteria, food antigens, and
immune cells [7-9]. Prior antibiotic use has been shown to deplete diversity of
commensal bacteria and to perturb the gut environment in ways that likely
increase the risk of growth of resistant bacteria [7-12]. Experimental studies
demonstrate that gut bacteria can develop resistance in as little as one day after
exposure to antibiotics, a rapid time course that may be explained by selective
pressure and horizontal transfer of resistant genes [8]. The degree to which the
gastrointestinal tract is exposed to antibiotics is likely to be influenced by the
route through which antibiotics are eliminated. The two major routes of

74

elimination are through the liver and biliary system, and through the kidneys.
Antibiotics eliminated through the liver and biliary system pass through the
gastrointestinal tract thereby exposing gut bacteria to antibiotics. Conversely, far
less exposure of gut bacteria to antibiotics is likely to occur when antibiotics are
eliminated through the kidneys. However, no epidemiologic studies have directly
assessed the impact of route of antibiotic elimination on the risk for subsequent
drug resistance in humans. Therefore, to test the hypothesis that the route of
antibiotic elimination is associated with subsequent drug resistance, we
conducted a retrospective hospital-based cohort study.
3.3 MATERIALS AND METHODS
We examined patterns of drug resistance of organisms cultured from the
bloodstream of patients with HA-BSI in relation to antibiotics used in treatment,
overall and by route of elimination. Participating patients had been admitted to
the ICU of Los Angeles County University of Southern California Medical Center
(LAC-USC MC), and individual patient-level data on antibiotic use and resistance
of organisms subsequently identified from blood cultures were obtained from
medical records. We predicted that among patients with HA-BSI, drug resistant
organisms would have been cultured more often from blood samples of those
treated with antibiotics than from samples of those with no antibiotics; we also
predicted that drug resistant organisms would have been cultured more often
from blood samples of those treated with antibiotics eliminated through the liver
and biliary system than from samples of those treated with antibiotics eliminated
through the kidneys. Finally, to assess specific patterns of antibiotic resistance

75

and organisms, we predicted that among patients with HA-BSI, there would have
been a greater frequency of emerging vancomycin-resistant Enterococci (VRE)
among patients administered vancomycin preceding the HA-BSI blood culture
compared to patients not administered vancomycin. We examined (1) whether
use of any antibiotic was associated with increased odds of subsequent drug
resistance of this form, (2) whether use of antibiotics eliminated at least partially
through the liver was associated with increased risk of subsequent drug
resistance of organisms identified from HA-BSIs, compared to use of antibiotics
eliminated predominantly through the kidneys, and (3) whether use of specific
antibiotics was associated with subsequent resistance of these antibiotics from
organisms identified from HA-BSIs. Finally, we estimated the proportion of
patients with drug resistant HA-BSI organisms for whom drug resistance was
attributable to antibiotic use.
Participants
In this retrospective cohort study, the association between antibiotic
elimination route and subsequent drug resistance was assessed among patients
admitted to the medical and surgical ICUs of LAC-USC MC between 1993 and
2003. Patients were classified according to the elimination route of antibiotics
received and degree of antibiotic resistance of HA-BSI organisms cultured at
least 48 hours after antibiotic administration. Patients were excluded if they were
less than 18 years old at admission, had an ICU identifier(s) that could not be
linked to the hospital identifier, or had missing data for antibiotic use. University
of Southern California Institutional Review Board approval was obtained under

76

protocol number HS-02B045 for analysis of these data from existing records with
no additional patient contact.
Data Collection
Data were collected for ICU patients in LAC-USC MC between 1993 and
2003 using the hospital Affinity Electronic Medical Record (EMR) database and a
separate ICU EMR database. A patient identifier for each patient was generated
using the hospital EMR at the time of their admission. Information collected
included inpatient admission date and time, admitting diagnosis, comorbidities,
admitting physician, inpatient bed assignment, date of birth, sex, race, weight,
surgeries and procedures, discharge disposition, discharge date, and
subsequent patient readmissions. Presence (yes/no) and type of antimicrobial
resistance (susceptible, intermediate, resistant) of cultured organisms were
collected for patients who had a positive blood culture in the hospital within the
study period. Admission into the ICU was captured separately by the ICU EMR
with a new ICU patient identifier for each admission into the ICU, including
readmissions. Date and time of ICU admission, sex, race, weight, pharmacy
data, mechanical ventilation, catheter use, blood transfusions, and laboratory
parameters by organ site were collected for all patients in the ICU.
The data were managed using Microsoft ACCESS and contained annual
medications datasets (date range: 1995-2003); annual laboratory datasets by
function (kidney function, liver function, hematology, blood pressure/pulse,
pancreatic function, diabetic control, disseminated intravascular coagulation, for
date range: 1994-2003); annual catheter datasets by type of line (arterial blood

77

gas, blood transfusion, central venous pressure, hemodialysis, pulmonary artery
pressure, total parenteral nutrition, ventilator, for date range: 1993-2003); a
dataset of microorganisms identified and susceptibility data (data range: 1994-
2003); admission, comorbidity, and discharge data in the hospital (date range:
1993-2003); and admission and discharge data in the ICU (date range: 1993-
2003). Data were extracted from the databases and linked using the hospital
patient identifier, ICU patient identifier, and hospital admission and discharge
date as keys.
BSIs were selected for the study because organisms cultured from blood
were likely to be pathogens responsible for the infection. Organisms cultured
from infected blood were less likely to have arisen from contamination of the
specimen than organisms cultured from other infected tissue compartments.
Blood samples submitted for cultures were collected for patients in the hospital
exhibiting clinical signs and symptoms of infection, beginning at the onset of
suspicion and repeated until negative at the discretion of the treating physician.
All blood cultures were processed within the LAC-USC MC Microbiology
Laboratory, and cultures were handled per the Los Angeles County Hospital
Microbiology Laboratory Bacteriology Culture Manual [13]. Positive blood
cultures were reported back through the EMR and susceptibility testing was
performed using a standard panel of antimicrobial agents based on gram stain
classification and classified as either susceptible, intermediate, or resistant to the
antimicrobial agent.

78

Data on all patients admitted to other units of the hospital and to the ICU
were merged with admission data, using hospital and ICU patient identifiers as
keys, to create the cohort of all patients in the ICU. Medication and
microorganism data were linked to all patients in the ICU using ICU and hospital
patient identifiers. Organisms were classified as being from hospital-acquired
bloodstream infections if the corresponding blood cultures were taken at least 3
days after admission to the hospital. We defined a minimum latency period of two
days between antibiotic exposure and the development of the bacterial
resistance profile of the cultured organism, as follows: among patients with both
antibiotic use and HA-BSI(s), only the resistance profile of organisms identified
from specimens collected at least two days after antibiotic exposure contributed
to the resistance score, and all antibiotics whose administration started at least
two days prior to blood culture sampling were used to determine the elimination
route. Admitting diagnoses into the hospital were classified by linking ICD-9-CM
diagnosis codes from admission data to ICD-9-CM diagnosis codes classified by
the Centers for Medicare and Medicaid Services (CMS) (Supplemental Table 1)
[14]. Selected baseline comorbidities were assessed for all patients in the cohort
and defined as the presence of comorbidities within a one year lookback period
from the earliest use of antibiotics in the ICU or blood culture accession date.
Comorbidities were identified using ICD-9-CM diagnosis codes and grouped
according to therapeutic area (Supplemental Table 2).
Classifying Route of Antibiotic Elimination

79

Antibiotics were identified from all medications administered to patients in
the ICU cohort. The antibiotic classification was reviewed independently by a
LAC-USC MC clinician and a LAC-USC MC pharmacologist. Antimicrobials other
than antibiotics were identified and reviewed, but were not used in this analysis.
Antibiotic elimination route was classified first for each antibiotic and then
summarized for each patient and organism. Elimination route was identified
based on whether dose adjustment of the antibiotic was required for patients with
decreased renal function, determined by creatinine clearance (Supplemental
Table 3) [15]. If no dose adjustment for decreased renal function for a specific
drug was necessary, the antibiotic was assumed to be partially or predominantly
eliminated through the liver. Otherwise, the antibiotic was assumed to be
eliminated predominately through the kidneys. Route of elimination of antibiotics
administered to individual patients was scored as either predominantly through
the kidneys or at least partially through the liver as follows. Patients who had
received at least one antibiotic eliminated at least partially through the liver up to
the time of the blood culture were classified as having received antibiotic
elimination through the liver. Those receiving only antibiotics eliminated
predominantly through the kidneys up to the time of the blood culture were
classified as having received antibiotic elimination through the kidneys.
Elimination through the liver was assumed to be cumulative and was carried
forward, so that patients receiving any antibiotics eliminated via this route prior to
the first blood culture were classified as having elimination through the liver for
the first and all subsequent blood cultures. Among patients with no blood cultures

80

or no organisms grown from blood cultures, the cumulative route of antibiotic
elimination was defined using all antibiotics administered during their ICU
experience. Patients with no antibiotic use during their ICU stay were classified
as having no antibiotic exposure.
Defining Organism Resistance
Microorganisms isolated from blood cultures were identified. Non-bacterial
microorganisms (virus, fungi) were excluded, and the remaining microorganisms
were reviewed by the study physician and pharmacist and were removed if they
were either ill-defined or resulting from probable contamination of specimen
collection. The following species were defined as likely contaminants and
removed: Bacillus cereus, Corynebacterium group G1, Corynebacterium group
G2, Corynebacterium jeikeium, Corynebacterium minutissimum,
Corynebacterium species, Corynebacterium xerosis, gram negative
Coccobacillus, coagulase negative staphylococcus, Staphylococcus epidermidis,
Staphylococcus sciuri. In addition, these results were not otherwise specified and
were removed: Bacillus species and Gram negative or positive rod. Finally,
records were removed if the gram stain classification was discordant with the
microorganism. Only HA-BSI blood cultures were considered in this analysis.
Organisms had been tested against a panel of antimicrobial agents to
determine resistance to each antimicrobial agent using standard microbiology
methods (broth microdilution and disk diffusion) according to CLSI and FDA
recommendations [16]. The specific set of antimicrobial agents against which an

81

organism was tested was determined by the gram stain classification of the
organism.
Organism resistance was derived using methods specified by the CDC
and ECDC [17]. Microorganisms were categorized into one of seven bacterial
groups based on resistance profile and gram stain classification. Gram negative
bacteria were categorized as Enterobacteriaceae family, Pseudomonas
aeruginosa, or Acinetobacter species, imputed as one of the three groups based
on their resistance profile, or classified as general gram negative [17]. Gram
positive isolates were categorized as Staphylococcus aureus or Enterococcus
species, imputed as either based on their resistance profile, or classified as
general gram positive [17]. The final mapping of all bacteria to one of the seven
groups underwent review by the study pharmacologist.
Drug resistance was scored using susceptibility data of each organism
from each isolate to the following antibiotic classes: aminoglycoside, rifamycins,
carbapenems, cephalosporins, fluoroquinolones, folate pathway inhibitors,
glycopeptides, lincosamides, macrolides, monobactams, nitrofuran,
nitroimidazoles, oxazolidinones, penicillins, phenicols, polymyxins,
streptogramins, sulfonamides, and tetracyclines. Resistance algorithms were
specific to each bacteria group, as activity of antimicrobial classes varies
between groups. Only data for antimicrobial classes with activity against the
specified bacterial group contributed to the resistance definition for each
organism. Bacteria with intrinsic resistance to an antibiotic or antimicrobial class,

82

such as Enterococcus faecium resistance to carbapenems, did not contribute to
the resistance derivation.
Organisms were categorized as non-resistant, multi-drug resistant (MDR),
extensive-drug resistant (XDR), or pan-drug resistant (PDR). MDR was defined
as resistance to at least one antimicrobial agent in at least three antimicrobial
classes; XDR was defined as resistance to at least one antimicrobial agent in all
but two or fewer antimicrobial classes; PDR was defined as resistance to all
antimicrobial agents in all classes. An organism was classified as non-resistant if
it was not MDR, XDR, and PDR. Patients with no cultured organisms were
classified as having no HA-BSI causative blood culture as the outcome. For
patients with a polymicrobial culture, the most resistant organism score was used
in all analyses. Patients with no organisms that met the HA-BSI criteria and
latency period criteria did not contribute to the resistance score and therefore
were classified as having no organism for the purposes of these analyses.
Statistical Analysis
Complete data from 1995 to 2002 for patients at least 18 years of age
were used in the analysis. Counts and proportions were presented for categorical
variables; mean, standard deviation, and range were presented for continuous
variables. Among patients with multiple blood cultures over time, the first
qualifying blood culture, defined as a blood draw from which a hospital-acquired
organism was cultured and after the latency period, was selected for analysis.
The independent variable was route of antibiotic resistance (no antibiotic use,
only antibiotics eliminated through the kidneys, any antibiotics eliminated through

83

the liver). The outcome was drug resistance (no organism, non-resistant
organism, MDR organism, XDR organism). The association between route of
antibiotic exposure and subsequent resistance was assessed for four sets of
patients: all patients in the ICU cohort, all patients in the ICU cohort with
antibiotic use, all patients in the ICU cohort with antibiotic use and a positive
blood culture, and all patients in the ICU cohort with ESKAPE bacteria. Route of
antibiotic elimination was modeled both as a categorical variable, and as a
continuous variable as an indicator for increasing antibiotic exposure in the gut
and assumed an equal increase in the beta estimates. Multinomial logistic
regression was used to estimate the association between the antibiotic
elimination route and subsequent drug resistance, modeling drug resistance as a
categorical variable with a static reference group. Among all patients in the ICU
and patients in the ICU with antibiotic use, the reference group was patients with
no organisms; among ICU patients with antibiotic use and a positive blood
culture, the reference group was patients with a non-resistant organism.
Cumulative logistic regression was used to model drug resistance on an ordinal
scale to assess the risk of increasing subsequent resistance. The Proportional
Odds assumption was tested and in those instances where the assumption did
not hold, partial proportional odds logistics regression was used. Logistic
regression was used to model antibiotic use for select organisms
The reference group of patients with no organism included patients with
no blood culture taken, patients for whom a blood draw cultured no organism,
and patients with only non-qualifying organisms. It is likely that patients with non-

84

qualifying organisms were more sick compared to patients with no blood cultures
and negative blood cultures; thus, sensitivity analyses were conducted for all
models presented whereby patients with non-qualifying organisms were removed
from the reference group in order to attempt to have patients of similar disease
severity in the baseline group. In all analyses, the presence of confounding by
age, sex, race, ICU category (Medical or Surgical), admission year, ICU LOS
(days), and specified comorbidities were assessed by adding covariates
individually into the model and evaluating the impact on the beta coefficient of the
main effects. Covariates were included in the model if their addition resulted in an
at least 15% change of the main effects. Effect modification by sex, race, gram
stain classification, and surgical versus medication ICU was assessed for all
analyses. Because the large proportion of untyped Staphylococcus species
found in our data may represent contamination during blood draws, additional
stratification of Staphylococcus species versus non-Staphylococcus species was
conducted for select analyses. The exposure attributable fraction (AF) was
calculated to determine the proportion of patients with antibiotic use and in the
ICU whose drug resistance could be attributable to antibiotic use. Odds ratios
were used to estimate effect size, using patients with no organism as the
baseline group. All statistical analyses were performed using SAS software,
Version 9.4 [Cary, NC].
3.4 RESULTS
Records were combined for hospital patients that had at least one ICU
stay (N=14,288) and those admitted only to the ICU (N=430). Patients with ICU

85

identifiers that could not be linked to hospital records (N=1,874 identifiers) were
excluded because resistance score could not be determined for these patients.
Of the 14,718 patients in the ICU, the 2,717 who had missing medication data
prior to September 1995, and the 425 who were younger than 18 years of age
were subsequently excluded (Figure 1). The final ICU cohort had 11,576
patients, of which 2,035 (17.6%) had received no antibiotic administrations, 5,176
(44.7%) had antibiotics only eliminated through the kidneys, and 4,365 (37.7%)
had received antibiotics eliminated at least partially through the liver.
Overall, 12,627 patients in the hospital had a positive blood culture at any
time during their hospital stay (Figure 1). A total of 191 different microorganisms
were identified from patients’ isolates, of which 15 organisms were classified as
contaminants or non-bacterial microorganism and were removed. The remaining
8,847 (70%) patients had qualifying microorganisms, and of these, 2,473 patients
were in the ICU. A further 969 patients (767 whose blood cultures did not meet
the HA-BSI criteria and 202 whose blood cultures met the HA-BSI criteria but did
not meet the latency criteria for antibiotic exposure) were removed from the
analysis, resulting in 1,079 patients in the ICU cohort with eligible blood cultures.
Overall, 10,484 (90.7%) patients in the ICU cohort had no HA-BSI, 566 (4.9%)
had a subsequent non-resistant HA-BSI, 398 (3.4%) had a subsequent MDR HA-
BSI, and 115 (1.0%) had a subsequent XDR HA-BSI. No patients had PDR
subsequent HA-BSI.
Among all patients in the ICU, the average (SD) age in the ICU was 47.5
(17.3) years and differed by antibiotic route (P<0.001) (Table 1). Approximately

86

67% of patients were male, 57.0% Hispanic, 15.4% White, 14.5% Black, and
9.5% Asian. Hispanics were not further classified separately by race; however,
for the purposes of this study, Hispanics were considered to be White. The
average hospital LOS (SD) was 23.5 days (38.4) and average ICU LOS was 8.5
days (11.4) (Table 2). Average ICU LOS (SD) varied by route of antibiotic
elimination (P<0.001). Patients with no antibiotic exposure had shortest average
ICU LOS (3.1 days, SD: 2.0) followed by patients who received antibiotics
eliminated through the kidneys (6.3 days, SD: 6.4). Patients with antibiotics
eliminated through the liver had the longest average ICU LOS (13.6 days, SD:
15.7). This same trend of increasing ICU LOS with increasing antibiotic exposure
to the gut was observed in patients admitted to all ICU types. ICU LOS differed
among the Surgical, Medical, and Neurosurgical ICU types (P=0.0045), although
it did not differ significantly between Medical and Surgical ICUs (P>0.05) (data
not shown). Patients in the Neurosurgical (NSICU) ICU had slightly shorter
average LOS compared to those in Surgical and Medical ICUs (7.4 days (SD:
6.8) compared to 8.6 days (SD: 12.8) and 8.6 days (SD: 10.7), respectively).
Admitting diagnosis categories varied by antibiotic exposure (P<0.0001)
(Table 1). Approximately half of all admitting diagnoses within the study period
were injury and poisoning conditions (26.7%) and symptoms, signs, and ill-
defined conditions (25.6%). The most common injuries were intracranial injuries,
traumatic complications and unspecified injuries, open wounds of the head, neck
and trunk, and fractures of spine and trunk, skull, and lower limb (data not
shown); the most common symptoms, signs, and ill-defined conditions were

87

general symptoms and symptoms involving the respiratory system and other
chest symptoms (data not shown). Patients with antibiotics eliminated through
the liver had the lowest proportion of injuries and poisoning and highest
proportion of symptoms, signs, and ill-defined conditions, compared to patients
with no antibiotics or patients with antibiotics eliminated through the kidneys.
Diseases of the digestive system accounted for 9.4% of all admitting diagnoses,
of which the most common conditions were gastrointestinal hemorrhage,
diseases of the pancreas, and chronic liver disease and cirrhosis. Overall, 8.9%
of patients in the ICU had diseases of the circulatory system upon admission, the
majority of which was cardiovascular disease. The proportion of diseases of the
cardiovascular system decreased with increasing antibiotic exposure to the gut.
Neoplasms constituted approximately 7.0% of admitting diagnoses and were
highest among patients with antibiotics eliminated through the kidneys. The most
common neoplasms were malignant neoplasms of the digestive organs and
peritoneum and of the genitourinary organs (data not shown). Diseases of the
respiratory system accounted for 5.2% of admitting diagnoses. Patients with
antibiotics eliminated through the liver had a higher percentage of diseases of the
respiratory system (9.4%), including pneumonia and influenza, compared to
patients with antibiotics eliminated through the kidneys or no antibiotics (2.8%
and 2.2%).
In the 1 year baseline period for patients in the ICU, cardiovascular
conditions accounted for 33.7% of comorbidities, followed by renal (16.6%),
cancer (13.7%), osteoporotic (12.1%), Type II diabetes (11.7%), gastrointestinal

88

(10.2%), and pulmonary (6.3%) conditions (Table 1). Renal, cancer,
gastrointestinal, osteoporotic, central nervous system, and pulmonary conditions
and Type II diabetes in the baseline period varied by antibiotic exposure
(P<0.05). The proportion of renal conditions, Type II diabetes, and pulmonary
conditions was highest for patients with antibiotics eliminated through the liver,
followed by patients with antibiotics eliminated through the kidneys, and was
lowest for patients with no antibiotic use. Patients with antibiotics eliminated
through the kidneys had higher proportions for osteoporotic conditions (16.1%
versus 7.4%) and cancer (16.6% versus 14.1%) compared to patients with
antibiotics eliminated through the liver. The most commonly prescribed antibiotics
eliminated predominantly through the kidneys were Cefazolin (23.5%),
Vancomycin (11.4%), Piperacillin and Tazobactam (11.3%), Levofloxacin
(11.1%), Gentamicin (10.0%), Cefotaxime (9.9%), and Ampicillin and Sulbactam
(5.2%) (Figure 2). Cefazolin, piperacillin and tazobactam, and vancomycin use
rose steadily to 11%, 15%, and 18%, with a percent change of 6%, 12%, and
9%, respectively, from the beginning to the end of study period (Figure 3).
Levofloxacin use at LAC-USC MC consistently rose from the approval date
(December 17, 1998) to greatest use of 13% in 2001, followed by tapered use
until the end of study period. Use of gentamicin and cefotaxime decreased by 9%
and 5% during the 8-year study period (Figure 3), while use of ampicillin and
sulbactam remained low and fairly steady during the study period. Among
patients who were administered any antibiotics eliminated through the liver, the
most commonly prescribed antibiotics were Vancomycin (12.3%), Gentamycin

89

(9.9%), Metronidazole (8.4%), Piperacillin and tazobactam (7.2%), Levofloxacin
(6.5%), and Clindamycin (5.6%) (data not shown). Of these patients, the most
commonly prescribed antibiotics with at least partial elimination through the liver
were Metronidazole (30.2%), Clindamycin (20.2%), Erythromycin (17.2%),
Oxacillin (11.0%), Ceftriaxone (7.9%), and Azithromycin (7.1%) (Figure 2).
In our study, 44.5% of the gram positive organisms cultured were untyped
Staphylococcus species (Figure 4). Among gram positive organisms, high
proportions of patients infected with Staphylococcus species, ranging from 71-
87%, were observed until 1998, after which these species were no longer
reported (Figure 5). This abrupt change may be due to a change in the assays
used for Staphylococcus species at the LAC-USC MC Microbiology Laboratory.
Proportion of the ESKAPE bacteria varied in our study. Staphylococcus aureus
was among the most commonly cultured (26.2%) gram positive bacteria, with
proportions peaking at 90% by 1998, decreasing to 38% by 2000 and
subsequently rising to 49% by the end of the study period. Enterococcus faecium
constituted a minor proportion (3.2%) of all gram positive organisms. It did not
appear in the LAC-USC MC ICU until 1999, with proportions between 4% and
14% until 2001, where it was no longer present. Of the gram negative ESKAPE
bacteria, 14.2% were Klebsiella pneumoniae, 12.8% were Pseudomonas
aeruginosa, 8.4% were Acinetobacter baumannii, and 20.4% were Enterobacter
(13.1% E. cloacae and 7.3% E. aerogenes) (Figure 4). Pseudomonas
aeruginosa, Enterobacter, and Klebsiella pneumoniae rates remained fairly
stable over time (Figure 5). Acinetobacter baumannii only appeared in the LAC-

90

USC MC ICU in 1999, and proportions of this bacteria slightly rose from 11% to
18% until the end of the study period. The percentage of HA-BSI causative blood
cultures differed by antibiotic elimination route (P<0.0001) (Table 3). More
patients with antibiotics eliminated through the liver had at least one HA-BSI
blood culture (14.5%) compared to patients with antibiotics eliminated through
the kidneys (7.0%). Only 4.1% of patients with no antibiotics had at least one
positive BSI. Approximately two-thirds of organisms cultured were gram positive.
This proportion varied by route of elimination (P<0.001): 58% in patients with
antibiotics eliminated through the kidneys and 75% in patients with no antibiotic
use.
A total of 11,563 patients with complete data were included in the
analyses; 13 patients were excluded from the analyses due to missing resistance
information. ICU LOS was a strong confounder of the route-resistance
association and kept in all models. Age was not a confounder but was
nevertheless kept in the model because of its clinical relevance. Patients with
any antibiotic use had higher odds of having drug resistance (OR: 2.59, 95% CI:
1.65-4.05). These patients had no increased odds of having a non-resistant
organism (P>0.05) (data not shown). Antibiotic use was associated with
increasing degree of resistance (P<0.0001). Patients with any antibiotic use were
more likely (OR: 2.83 [95% CI: 1.81, 4.43], P<0.0001) to develop resistance
(multi- or extensive-drug resistance) and extensive-drug resistance (OR: 6.49
[95%: 1.59, 26.61], P=0.0093) versus less resistance, compared to patients not
administered any antibiotics (data not shown).

91

Estimated associations between antibiotic elimination route and degree of
resistant bacteremia are shown in Table 4a. Any antibiotic use, regardless of
elimination route, was associated with elevated risk of both multi-drug resistance
(OR: 2.31, 95% CI: 1.44, 3.70) and extensive-drug resistance (OR: 5.32, 95% CI:
1.30, 21.76). Patients receiving antibiotics eliminated predominately through the
kidneys had an elevated risk of both multi-drug resistance and extensive-drug
resistance (OR: 1.82, 95% CI: 1.11, 2.97 and OR: 3.46, 95% CI: 0.82, 14.68
respectively), although this effect is not significant among the extensive-drug
resistance group (P=0.0923). Patients with any antibiotic eliminated at least
partially through the liver were more likely to have a multi-resistance (OR: 2.99,
95% CI: 1.84, 4.85) and extensive-drug resistant (OR: 7.60, 95% CI: 1.84, 31.41)
bacteremia compared to patients receiving no antibiotics. As expected, sensitivity
analysis removing patients with non-eligible organisms in the no organism
reference group had stronger effect estimates throughout (Supplementary Table
4), while resulting in the same trends as noted above. Stronger effect estimates
were observed for all subsequent stratified analyses, regardless of how the data
were modeled (Supplementary Tables 5, 7, 8).
ESKAPE organisms (N=412) accounted for 38.2% of all qualifying
organisms identified in our study. We observed elevated odds (OR=2.01, 95%
CI: 1.05, 4.16, P=0.0349) of resistance of ESKAPE bacteria for patients with
antibiotic use compared to patients with no antibiotic use (Table 4b). Patients
with antibiotics eliminated through the kidneys had elevated odds of having
resistant ESKAPE bacteria (OR=1.80 [95% CI: 0.88, 3.66]), although this was not

92

significant (P=0.1066). Patients with antibiotics eliminated through the liver had
2.49 (95% CI: 1.22, 5.05) times the odds of resistant ESKAPE bacteria compared
to patients with no antibiotic administration (P=0.0119). Among patients with only
ESKAPE bacteria, those with antibiotics eliminated through the liver had elevated
odds of resistance (OR=2.68 [95% CI: 1.19, 6.04]) compared to patients with no
antibiotics (P=0.0200) (data not shown). Furthermore, increasing exposure of
antibiotics in gut increases odds of resistance of ESKAPE bacteria (P=0.0193)
(data not shown). Due to zero cells, we were unable to determine effect
estimates for multi-drug and extensive drug resistance of ESKAPE bacteria.
Overall and by the route of antibiotic elimination, Whites appeared to have
higher magnitude of MDR risk and lower XDR risk compared to other
racial/ethnic groups, although differences in effect estimates were not statistically
significant and confidence bounds were unstable due to small sample sizes
(Supplemental Table 9). Asians had increased odds of having a non-resistant
organism, both overall and by route of antibiotic elimination. The association
between elimination route and resistance did not appear to be modified by
gender (Supplemental Table 10); however, compared to males, females had a
higher odds of having a non-resistant organism regardless of the route of
antibiotic elimination. In the subset of patients with a positive blood culture
(N=1,079), we observed increased magnitude of MDR risk among patients with a
gram negative organism cultured, compared to those with a gram positive
organism; this pattern was noted regardless of elimination route, although not

93

statistically significant (Supplemental Table 11). Zero cell count precluded
calculation of XDR risk for patients with gram negative organisms.
To utilize the inherent ordering of the resistance variable (decreasing to
increasing resistance), we estimated associations between the route of antibiotic
elimination and resistance by modeling resistance on the ordinal scale (Table 5).
Patients with antibiotics eliminated only through the kidneys had increasing odds
of developing more resistant subsequent infections: the odds of developing an
MDR or XDR infection versus a non-resistant infection or no infection were 2.08
(95% CI: 1.31, 3.31) (P=0.0020) while odds of developing an XDR infection
versus a less resistant infection or no infection were 3.93 (P=0.0636). Stronger
associations were observed for patients with antibiotics eliminated through the
liver: the odds of having at least an MDR infection versus less and XDR infection
versus less was 3.84 (95% CI: 2.43, 6.06) and 9.72 (95% CI: 2.35, 40.24).
Sensitivity analysis removing patients with non-eligible organisms from the group
of patients with no organisms showed the same patterns of resistance across all
groups of patients with antibiotic use but with stronger effect estimates
(Supplementary Tables 5). We also modeled the exposure continuously as an
indicator of antibiotic exposure in the gut, assuming an equal change in effect
size for each increase in antibiotic use (Supplementary Table 6). Type of
antibiotic used (equal increases in level specified as going from no antibiotic to
antibiotic eliminated through the kidneys, or from antibiotics eliminated through
the kidneys to those eliminated through the liver) was associated with increasing
resistance (P<0.0001). The odds of detecting an organism increased by 33%

94

(95% CI: 19%, 48%), the odds of having a multi-drug resistant or extensive-drug
resistant organism versus less resistance increased by 90% (95% CI: 60%,
125%), and the odds of having an extensive-drug resistant organism versus less
resistance increased by 169% (95% CI: 79%, 304%) for increasing gut exposure
to antibiotic use. Compared to having no organism, increasing antibiotic use by
one level increased the risk of detecting an organism by 4% (95% CI: -10%,
19%); increased the odds of having a multi-drug resistant organism by 69% (95%
CI: 40%, 140%); and increased the odds of having an extensive-drug resistance
organism by 138% (58%, 257%).
We further interrogated the relationship between antibiotic use and
subsequent resistance by conducting analyses in the subset of ICU patients with
antibiotic use (N=9,528) (Table 6). Elimination through the liver was associated
with increased odds of any subsequent drug resistance (P<0.0001) (data not
shown). The odds of patients developing MDR and XDR were 65% (95% CI:
30%, 111%) and 122% (95% CI: 37%, 259%) greater, respectively, for those
receiving antibiotics eliminated through the liver compared to those receiving
antibiotic eliminated through the kidneys. Elimination through the liver increased
the odds of more severe resistance compared to elimination through the kidneys
(P<0.0001). The odds of detecting an organism of any resistance were 32%
(95% CI: 13%, 53%) greater, the odds of having a multi-drug resistant or
extensive-drug resistant organism versus less resistance were 86% (95% CI:
49%, 131%) greater, and the odds of having an extensive-drug resistant
organism versus less resistance were 150% (95% CI: 54%, 306%) greater for

95

patients with any antibiotics eliminated though the liver compared to those with
only antibiotics eliminated through the kidneys.
In analyses further limited to patients with antibiotic use and a positive
blood culture (N=995), antibiotics eliminated through the liver were again found to
be associated with subsequent resistance (Supplemental Table 12). Patients with
antibiotics eliminated through the liver had 1.64 (95% CI: 1.21, 2.21) (P=0.0013)
times the odds of having an organism that was multidrug resistant versus non-
resistant and 2.17 (95% CI: 1.30, 3.62) (P=0.0029) times the odds of having an
organism that was extensive-drug resistant versus non-resistant. Patients had a
72% increase in odds of increasing resistance (95% CI: 31%, 126%) versus less
resistance compared to those who had only antibiotics eliminated through the
kidneys. Among patients who cultured non-Staphylococcus species, the odds of
developing multidrug resistance and extensive-drug resistance was 1.39 (95%
CI: 0.95, 2.03) and 1.69 (95% CI: 0.89, 3.20) for patients with antibiotics
eliminated through the liver compared to through the kidneys. Patients had a
44% increase in odds of increasing resistance (95% CI: 3%, 103%) versus less
resistance compared to those who had only antibiotics eliminated through the
kidneys. Among patients with antibiotic use and positive blood cultures,
proportions of non-resistance initially decreased but increased towards the end of
the study period (Figure 6). Rates of multidrug resistance followed the opposite
trend, while extensive-drug resistance were fairly stable over time.
Among patients with Enterococci, we found that patients with
administration of vancomycin preceding the blood culture had higher odds of

96

vancomycin resistance Enterococci (VRE) (OR=4.93 [95% CI: 2.16, 11.24])
compared to patients with no vancomycin administration (P<0.0001) (Table 7).
The association was not confounding by any of the covariates previously
considered, but was modified by the presence of a cardiac event in the baseline
period. For patients with no cardiac event, patients with vancomycin were 12.55
(95% CI: 3.74, 42.14) times as likely to have VRE in the blood (P<0.0001); for
patients with a cardiac event, those who had vancomycin were 1.71 (95% CI:
0.52, 5.45) times as likely to have VRE in the blood compared to patients with no
did not have vancomycin (P=0.3792).
Among patients with any antibiotic use during the ICU stay, 61.3% (95%
CI: 39.5%, 75.3%) of subsequent drug resistance was attributable to the
antibiotics used (Table 8). A higher percentage of drug resistance was
attributable to antibiotic administered through the liver compared to through the
kidneys (72.8% versus 41.7%, respectively). This pattern was also observed by
type of resistance: patients with any antibiotics eliminated through the liver had
elevated proportions (68.6%, 95% CI: 95% CI: 48.9%, 80.7%) of MDR resistance
attributable to their antibiotics compared to patients with only antibiotics
eliminated through the kidneys (40.3%, 95% CI: 1.2%, 63.9%), although the
confidence intervals overlapped. Similarly, patients administered any antibiotics
eliminated through the liver had higher proportions (88.3% vs 57.2%) of XDR
resistance attributable to antibiotics compared to patients with administered
antibiotics eliminated only through the kidneys. Sensitivity analyses removing
patients with no qualifying organism from the baseline group used to estimate

97

risk estimates for the exposure attributable fraction showed similar but stronger
proportions throughout (Supplemental Table 13). Finally, among patients with
any antibiotic use during the ICU stay, 52.2% (95% CI: 5.12%, 76.0%) of
subsequent drug resistance to ESKAPE organisms was attributable to the
antibiotics used (Table 9). Patients with antibiotics eliminated through the liver
had 63.1% (95% CI: 24.9%, 81.9%) of resistance to ESKAPE organisms
attributable to their antibiotic use.
3.5 DISCUSSION
Our analysis of N=11,573 patients which tested the association between
route of antibiotic elimination (no antibiotic use, antibiotics eliminated only
through the kidneys, and any antibiotics eliminated through the liver) and degree
of drug resistance showed that use of any antibiotic in the ICU setting increased
odds of drug resistance in subsequent HA-BSIs. This general finding accords
with earlier research, but earlier authors had not addressed the hypothesis that
antibiotic use in the ICU is associated with drug resistance of subsequent HA-BSI
organisms, or distinguished between antibiotics based on the route of elimination
and drug resistance of subsequent HA-BSI organisms. Our study found that
exposure to antibiotics that involve elimination through the gut increased the
odds of developing more resistant bacteremia for both multi-drug resistant and
extensive-drug resistant bloodstream infections. We also found that a large
proportion of subsequent drug resistance among patients with any antibiotic use
was due to the antibiotic (AF=61.3%, Table 7), and this was higher for antibiotics

98

eliminated through the liver compared to antibiotics eliminated through the
kidneys.
Our findings are consistent with those of previous studies of the
distribution of gram positive and gram negative BSIs in the hospital in that the
majority (66.8%) of qualifying organisms identified in our study were gram
positive (Table 3, Figure 4) [2]. All of the six ESKAPE organisms were identified
in the ICU population of our study and represented 38% of all organisms in the
primary analysis (Table 3). Rates of other common organisms identified in the
ICU were comparable to another hospital-based cohort study: Escherichia coli
was found in 6.8% of the cultures, and Staphylococcus species was found in
29.7% in our study compared to 6% and 31% [2]. Previous studies had already
identified prior antibiotic use as a risk factor of emergence of organism resistance
[10, 11]. Bodro, et al. found that cancer patients with prior antibiotic use were
1.53 (95% CI: 1.1, 2.2) times as likely to have a resistant ESKAPE organism, and
Vazquez, et al. found that risk of resistance varied by antibiotic and ranged from
1.54 to 2.13 for patients with prior antibiotic use [10, 11]. This is consistent with
findings from our study, which found that patients with prior antibiotic use were
1.87 (95% CI: 1.28, 2.72) times as likely to have a resistant organism as people
with no antibiotic use. In our study, no patients were identified as having pan-
drug resistance. This may be because our study period precedes the occurrence
of PDR organisms in the hospital. The greater magnitude of our effect estimates
compared to those in Bodro, et al. may be because resistance outcomes among

99

all organisms are included in this study but limited to the ESKAPE organisms in
Bodro, et al.
Although this is, to our knowledge, the first study to assess the association
between antibiotic elimination route and degree of organism resistance among a
large cohort of hospital patients, a few studies have examined the relationship
between the antibiotic pharmacologic properties and resistance in animal models
[18, 19]. One study found that antibiotic administration route was associated with
both the antibiotic resistance gene pool and antibiotic resistant bacteria in feces
of mice inoculated with non-resistant Enterococcus species or Escherichia coli
[18]. This study found that orally administered ampicillin conferred greater
resistance compared to intravenously (IV) administered ampicillin, while both
orally and IV administered tetracycline conferred greater resistance. In light of
our results, it seems plausible that this difference is due to the route of
eliminations, since IV ampicillin is excreted primarily through the kidneys
whereas IV tetracycline is excreted both through the kidneys and liver [15].
Several strengths can be noted for our study. Apart from being the first
epidemiologic study to address the association between elimination routes and
resistance, our study has used a clearly defined exposure that is based on
objective data. It uses a large hospital-based dataset which spans 8 years, and
can be generalized to the ICU patients of hospitals in the US with similar history
of antibiotic use and disease severity.
One limitation of our study is use of a categorical variable broadly
classifying route of antibiotic elimination into “through the kidneys” or “through the

100

liver” categories, rather than a continuous variable quantifying gut exposure to
antibiotics. However, such misclassification of the exposure is likely to be non-
differential, and expected to bias estimates of the association of route of
antibiotic elimination and drug resistance towards the null value. Our definition of
exposure would not explain the strong association between elimination route and
drug resistance found in this study. Although pharmacokinetic studies can be
used to determine percent renal or hepatic elimination, such data are not
available or incomplete for less-commonly used drugs and imprecisely describe
elimination rates. Moreover, compared to measures of antibiotic excretion in
urine, measure of excretion into bile and feces are more technically demanding,
so elimination by this route may be underreported [20]. Thus, classifying
elimination on a continuous scale faces numerous challenges. A further limitation
of our data is that because multiple antibiotics were administered to some
patients, we were unable to determine which antibiotics were the biggest culprit
in bacteremia resistance. Although we cannot rule out the possibility of influence
by an unrecognized form of systematic error, the pattern of higher odds of
resistance among patients with any antibiotics eliminated only through the liver
was apparent in analyses of multiple subsets of data using alternative reference
groups and survived numerous sensitivity analyses.
Another limitation of our study was that the reference group of patients
with no organism was comprised of patients with no blood culture taken, patients
with a blood culture that grew no organism, and patients with a positive blood
culture that had no qualifying organism to be considered for resistance scores. Of

101

the last group of patients, 79% had prevalent BSIs and the remaining 21% had
BSIs whose drug resistance was likely not be caused by the antibiotics they used
because the latency criteria was not met. Thus, it is likely that this group of
patients were more sick compared to patients with no blood culture or negative
blood culture, whom we presume to be less sick. Therefore, we predicted that the
effect estimates of the route-resistance association would be underestimated
when combining these three groups of patients in the reference group. Indeed,
sensitivity analyses removing patients with no qualifying organisms from the
reference group showed increased effect estimates of the association between
the route of antibiotic elimination and drug resistance for all subsets of data.
Finally, data from our study indicated that untyped Staphylococcus
species constituted 44.5% of all gram positive bacteria and 29.7% of all
qualifying organisms. Staphylococcus species were only present in our data until
1998, after which they were no longer found until the end of the study period.
One possible explanation for the sudden drop of Staphylococcus species from
our data is a change in the assay used to identify this species at the LAC-USC
MC Microbiology Laboratory. It is likely that the majority of the untyped
Staphylococcus species were coagulase-negative Staphylococcus species
(CoNS) and thus have been thought to be either a source of contamination or
less pathogenic infection compared to Staphylococcus aureus. True rates of
contamination from CoNS are unknown. Various studies have classified probable
contamination as ranging from 15% to 64% using Laboratory Confirmed
Bloodstream Infection guidelines by the CDC to determine probable infection [21-

102

23]; however, recent data have shown that CoNS contribute to BSIs by way of
foreign-body related infections, especially among patients that are
immunocompromised [24-26], and that rates of non-contaminated CoNS are high
among patients with foreign bodies [25]. Therefore it is difficult to determine the
proportion of untyped Staphylococcus species in our study that are contaminants
versus probable agents of infection. If contaminants, then the positive association
found between the route of elimination and Staphylococcus species
(Supplemental Table 8) may indicate possible confounding by indication for
patients administered antibiotics eliminated through the kidneys compared those
administered antibiotics through the kidneys. Methods to discriminate probable
contamination of the CoNS specimen from clinically relevant CoNS specimens
involve either correlation of the specimen with patient signs and symptoms of
infection or the use of a secondary blood culture to verify presence of the same
species, and can be addressed in a future study.
In conclusion, this study identified a robust pattern of resistance predicted
by the hypothesis that patients administered any antibiotics eliminated at least
partially through the liver will have increased drug resistance of subsequent HA-
BSI organisms. If further research substantiates antibiotic elimination influences
the risk for development of drug resistance, then limiting hepatic elimination of
antibiotic compounds may be a novel strategy with substantial potential to
decelerate emergence of antibiotic resistance, guiding infectious disease
practices of selecting among available antibiotics, and rational development of
new antibiotics in the longer term.

103

3.6 TABLES AND FIGURES
Table 3.1. Baseline and Demographic Characteristics of LA-USC MC ICU
Patients, 1995-2002

All Patients
(N=11,576)
No Antibiotic
Exposure
(N=2,035)
Through the
Kidneys
(N=5,176)
Through the
Liver
(N=4,365)

Patient Characteristics
Age
(years)**
N 11,576 2,035 5,176 4,365
Mean (SD) 47.5 (17.3) 47.1 (17.6) 47.0 (17.7) 48.3 (16.6)
Median 47.0 46.0 47.0 47.0
25th, 75th
percentile
34.0, 60.0 33.0, 59.0 32.0, 60.0 36.0, 60.0

Sex
Female n (%) 3,862 (33.4%) 651 (32.1%) 1,793 (34.6%) 1,418 (32.5%)
Male n (%) 7,705 (66.6%) 1,379 (67.9%) 3,380 (65.3%) 2,946 (67.5%)

Race/Ethnicity**
Asian n (%) 1,070 ( 9.5%) 205 (10.5%) 502 (10.0%) 363 ( 8.5%)
American
Indian
n (%) 13 ( 0.1%) 2 ( 0.1%) 5 ( 0.1%) 6 ( 0.1%)
Black n (%) 1,629 (14.5%) 297 (15.2%) 645 (12.8%) 687 (16.2%)
Hispanic n (%) 6,398 (57.0%) 1,052 (54.0%) 2,963 (58.9%) 2,383 (56.1%)
White n (%) 1,731 (15.4%) 299 (15.3%) 740 (14.7%) 692 (16.3%)
Other n (%) 165 ( 1.5%) 37 ( 1.9%) 75 ( 1.5%) 53 ( 1.2%)
Unknown n (%) 216 ( 1.9%) 56 ( 2.9%) 97 ( 1.9%) 63 ( 1.5%)

Weight (lbs)* N 11,237 1,914 5,006 4,317
Mean (SD) 165.4 (46.4) 163.4 (42.6) 166.9 (46.8) 164.5 (47.4)
Median 159.9 159.2 160.1 157.9
25th, 75th
percentile
135.8, 186.1 135.1, 183.4 138.0, 187.6 133.6, 185.0

Admitting Diagnosis Categories
Injury and
Poisoning
n (%) 2,976 ( 26.7%) 622 ( 32.1%) 1,625 ( 32.5%) 729 ( 17.3%)
Symptoms, signs,
and ill-defined
conditions
n (%) 2,862 ( 25.6%) 444 ( 22.9%) 1,072 ( 21.4%) 1,346 ( 31.9%)
Diseases of the
digestive system
n (%) 1,048 ( 9.4%) 162 ( 8.4%) 396 ( 7.9%) 490 ( 11.6%)
Diseases of the
circulatory system
n (%) 990 ( 8.9%) 276 ( 14.2%) 477 ( 9.5%) 237 ( 5.6%)
Neoplasms n (%) 778 ( 7.0%) 71 ( 3.7%) 466 ( 9.3%) 241 ( 5.7%)
Diseases of the
respiratory system
n (%) 580 ( 5.2%) 43 ( 2.2%) 140 ( 2.8%) 397 ( 9.4%)

Two patients with “unknown” and “other” sex were classified as having missing sex.
* Significant at the 0.05 alpha level
** Significant at the 0.001 alpha level

104

Table 3.1. continued

All Patients
(N=11,576)
No Antibiotic
Exposure
(N=2,035)
Through the
Kidneys
(N=5,176)
Through the
Liver
(N=4,365)

Select Comorbidities in Baseline Period
Renal** n (%) 1,918 ( 16.6%) 198 ( 9.7%) 663 ( 12.8%) 1,057 ( 24.2%)
Cancer** n (%) 1,590 ( 13.7%) 114 ( 5.6%) 860 ( 16.6%) 616 ( 14.1%)
Type II Diabetes* n (%) 1,352 ( 11.7%) 216 ( 10.6%) 571 ( 11%) 565 ( 12.9%)
Gastrointestinal** n (%) 1,182 ( 10.2%) 222 ( 10.9%) 366 ( 7.1%) 594 ( 13.6%)
Cardiovascular n (%) 3,898 ( 33.7%) 706 ( 34.7%) 1,665 ( 32.2%) 1,527 ( 35%)
Osteoporotic** n (%) 1,395 ( 12.1%) 239 ( 11.7%) 833 ( 16.1%) 323 ( 7.4%)
CNS* n (%) 113 ( 1%) 19 ( 0.9%) 34 ( 0.7%) 60 ( 1.4%)
Pulmonary** n (%) 724 ( 6.3%) 89 ( 4.4%) 238 ( 4.6%) 397 ( 9.1%)
Connective Tissue
Disorders
n (%)
131 ( 1.1%) 23 ( 1.1%) 47 ( 0.9%) 61 ( 1.4%)

* Significant at the 0.05 alpha level
** Significant at the 0.001 alpha level

105

Table 3.2. Hospitalization Characteristics of LA-USC MC ICU Patients, 1995-
2002

All Patients
(N=11,576)
No Antibiotic
Exposure
(N=2,035)
Through the
Kidneys
(N=5,176)
Through the
Liver
(N=4,365)

ICU Category**
SICU n (%) 4,981 (43.1%) 730 (35.9%) 2,840 (54.9%) 1,411 (32.4%)
MICU n (%) 5,444 (47.1%) 1,075 (52.8%) 1,612 (31.2%) 2,757 (63.3%)
NSICU n (%) 1,137 ( 9.8%) 230 (11.3%) 717 (13.9%) 190 ( 4.4%)

Days in Hospital**
N 11,249 1,949 5,039 4,261
Mean (SD) 23.5 (38.4) 9.1 (9.8) 19.6 (26.4) 34.7 (52.9)
Median 14.0 6.0 13.0 21.0
25th, 75th
percentile
7.0, 26.0 4.0, 11.0 8.0, 23.0 11.0, 39.0

Days in ICU**
N 11,576 2,035 5,176 4,365
Mean (SD) 8.5 (11.4) 3.1 (2.0) 6.3 (6.4) 13.6 (15.7)
Median 5.0 3.0 4.0 8.0
25th, 75th
percentile
3.0, 9.0 2.0, 4.0 3.0, 7.0 4.0, 17.0

Days in Surgical ICU**
N 4,981 730 2,840 1,411
Mean (SD) 8.6 (12.8) 3.0 (1.9) 6.0 (6.6) 16.6 (20.0)
Median 4.0 2.0 4.0 9.0
25th, 75th
percentile
3.0, 8.0 2.0, 4.0 3.0, 7.0 4.0, 22.0

Days in Medical ICU**
N 5,444 1,075 1,612 2,757
Mean (SD) 8.6 (10.7) 3.1 (2.0) 6.4 (6.4) 12.1 (13.2)
Median 5.0 3.0 4.0 8.0
25th, 75th
percentile
3.0, 10.0 2.0, 4.0 3.0, 8.0 4.0, 15.0

Days in Neurosurgical ICU**
N 1,137 230 717 190
Mean (SD) 7.4 (6.8) 3.8 (2.6) 7.1 (5.7) 12.9 (10.1)
Median 5.0 3.0 5.0 11.0
25th, 75th
percentile
3.0, 10.0 2.0, 4.0 3.0, 10.0 6.0, 17.0

SICU=Surgical ICU, MICU=Medical ICU, NSICU=Neurosurgical ICU.
* Significant at the 0.05 alpha level
** Significant at the 0.001 alpha level

106

Table 3.3. Microbiological Characteristics of LA-USC MC ICU Patients, 1995-
2002

All Patients
(N=11,576)
No Antibiotic
Exposure
(N=2,035)
Through the
Kidneys
(N=5,176)
Through the
Liver
(N=4,365)

Patients with at least
one positive blood
culture
a,b
**
N1 (%) 1,079 (9.3%) 84 (4.1%) 362 (7.0%) 633 (14.5%)

Gram Stain
Classification
c
**

Gram negative n (%) 358 (33.2%) 21 (25.0%) 151 (41.7%) 186 (29.4%)
Gram positive n (%) 721 (66.8%) 63 (75.0%) 211 (58.3%) 447 (70.6%)

Microorganisms
c
**
Acinetobacter
baumannii
n (%) 30 (2.8%) 0 (0.0%) 10 (2.8%) 21 (3.3%)
Staphylococcus aureus n (%) 189 (17.5%) 37 (44.0%) 57 (15.8%) 97 (15.3%)
Klebsiella pneumoniae n (%) 51 (4.7%) 5 (6.0%) 22 (6.1%) 24 (3.8%)
Pseudomonas
aeruginosa
n (%) 46 (4.3%) 4 (4.8%) 19 (5.3%) 24 (3.8%)
Enterobacter species
d
n (%) 73 (6.8%) 4 (4.8%) 34 (23.2%) 36 (5.7%)
Enterococcus faecium n (%) 23 (2.1%) 1 (1.2%) 4 (1.1%) 18 (2.8%)
Escherichia coli n (%) 73 (6.8%) 7 (7.1%) 40 (10.8%) 28 (4.4%)
Staphylococcus species n (%) 321 (29.7%) 21 (25.0%) 75 (20.7%) 232 (36.7%)

a
Inclusive of all organisms meeting the eligibility criteria for the study.
b
Percentage for patients with at least one positive blood culture is out of all ICU patients (N1/N).
c
Percentage for organisms is out of number of patients with at least one positive blood culture
(n/N1).
d
Enterobacter species includes Enterobacter cloacae and Enterobacter aerogenes.
* Significant at the 0.05 alpha level using Chi-squared test.
** Significant at the 0.001 alpha level using Chi-squared test.

107

Table 3.4a. Association between Antibiotic Elimination Route and Drug
Resistance of HA-BSI Organisms in ICU Patients using Multinomial Model, 1995-
2002

Resistance

Route
a,b

No
Organism
Non-resistant
Organism
MDR
Organism
†

XDR
Organism
††

OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI)

No antibiotic use
N 1,951 63 19 2

Any antibiotic use
N 8,533 503 379 113
Odds 1.0 1.12 (0.85, 1.48) 2.31 (1.44, 3.70) 5.32 (1.30, 21.76)
p-value -- 0.4136 0.0005 0.0201

Through the
kidneys

N 4,808 225 113 24
Odds 1.0 1.15 (0.86, 1.53) 1.82 (1.11, 2.97) 3.46 (0.82, 14.68)
p-value -- 0.3510 0.0169 0.0923

Through the liver
N 3,725 278 266 89
Odds 1.0 1.12 (0.83, 1.51) 2.99 (1.84, 4.85) 7.60 (1.84, 31.41)
p-value -- 0.4625 <0.0001 0.0051

a
N=11,563 patients were included in the analysis; 13 patients were excluded due to missing
resistant scores.
b
Odds ratios are reported from multinomial logistic regression adjusted for age and ICU LOS
using no organism as the reference group for resistance and no antibiotics as the reference
group for elimination route.
†
Multi-drug resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

108

Table 3.4b. Association between Antibiotic Elimination Route and Drug
Resistance of HA-BSI ESKAPE Organisms in ICU Patients using Multinomial
Model, 1995-2002

Resistance

Route
a,b

No
Organism
Non-resistant
Organism
MDR or XDR
Organism
†

OR (95% CI) OR (95% CI) OR (95% CI)

No antibiotic use
N 1,951 40 9

Any antibiotic use
N 8,533 204 159
Odds 1.0 0.80 (0.56, 1.14) 2.09 (1.05, 4.16)
p-value -- 0.2119 0.0349

Through the kidneys
N 4,808 92 52
Odds 1.0 0.78 (0.53, 1.14) 1.80 (0.88, 3.66)
p-value -- 0.1963 0.1066

Through the liver
N 3,725 112 107
Odds 1.0 0.83 (0.56, 1.23) 2.49 (1.22, 5.05)
p-value -- 0.3548 0.0119

a
N=10,896 patients were included in the analysis; 13 patients were excluded due to missing
resistant scores.
b
Odds ratios are reported from multinomial logistic regression adjusted for age and ICU LOS
using no organism as the reference group for resistance and no antibiotics as the reference
group for elimination route.
†
Multi-drug resistance (MDR) defined as resistance to ≥3 antimicrobial classes. Extensive-drug
resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

109

Table 3.5. Association between Antibiotic Elimination Route and Drug Resistance
of HA-BSI Organisms in ICU Patients using Cumulative Model, 1995-2002
No
Organism
Non-Resistant
Organism
MDR
Organism
†
XDR
Organism
††

No antibiotic use
N 1,951 63 19 2

Any antibiotic use
N 8,533 503 379 113
OR
a
(95% CI) 1.0 1.55 (1.23, 1.96) 2.83 (1.81, 4.43) 6.49 (1.59, 26.61)
p-value -- 0.0002 <0.0001 0.0093

Through the
kidneys

N 4,808 225 113 24
OR
a
(95% CI) 1.0 1.38 (1.08, 1.76) 2.08 (1.31, 3.31) 3.93 (0.93, 16.66)
p-value -- 0.0106 0.0020 0.0636

Through the liver
N 3,725 278 266 89
OR
a
(95% CI) 1.0 1.80 (1.40, 2.31) 3.84 (2.43, 6.06) 9.72 (2.35, 40.24)
p-value -- <0.0001 <0.0001 0.0017

N=11,563 patients were included in the analysis; 13 patients were excluded due to missing
resistant scores.
a
Odds ratios are reported from partial proportional odds model adjusted for age and duration
in the ICU. Resistance gradient is from less severe to more severe antibiotic resistance.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

110

Table 3.6. Association between Antibiotic Elimination Route and Drug Resistance
of HA-BSI Organisms Among ICU Patients with Antibiotic Use, 1995-2002

Resistance
No
Organism
Non-resistant
Organism
Organism with
MDR
†
Organism with
XDR
††

Through the kidneys
N 4,808 225 113 24

Through the liver

N 3,725 278 266 89
Multinomial Model
a

OR (95% CI) 1.0 0.99 (0.81, 1.20) 1.65 (1.30, 2.11) 2.22 (1.37, 3.59)
p-value -- 0.8923 <0.0001 0.0012
Cumulative Model
b

OR (95% CI) 1.0 1.32 (1.13, 1.53) 1.86 (1.49, 2.31) 2.50 (1.54, 4.06)
p-value -- 0.0003 <0.0001 0.0002

N=9,528 patients were included in the analysis; 13 patients were excluded due to missing
resistant scores.
a
Odds ratios are reported from multinomial logistic regression model adjusted for age and
duration in the ICU using no organism as the reference group for resistance score.
b
Odds ratios are reported from partial proportional odds model adjusted for age and duration in
the ICU, and represent the odds of resistance of the observed level or greater, versus
everything less severe.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

111

Table 3.7. Antibiotic-Organism Associations of HA-BSI Organisms in ICU
Patients using Multinomial Model, 1995-2002

Resistance
Organism Susceptible Resistant

Enterococci
Overall
No Vancomycin use
N 82 13
Vancomycin use
N 28 21
OR
a
(95% CI) 1.0 4.93 (2.16, 11.24)
p-value -- <0.0001

No Cardiac Event
No Vancomycin use
N 57 5
Vancomycin use
N 12 13
OR
a
(95% CI) 1.0 12.55 (3.74, 42.14)
p-value -- <0.0001

Cardiac Event
No Vancomycin use
N 25 8
Vancomycin use
N 16 8
OR
a
(95% CI) 1.0 1.71 (0.52, 5.45)
p-value -- 0.3792

N=144 patients were included in the analysis. Resistance or susceptibility of Enterococci to
Vancomycin are based on standard microbiology results provided in the susceptibility data.
a
Odds ratios are reported from multinomial logistic regression adjusted for age.

112

Table 3.8. Fraction of Drug Resistance Attributable to Antibiotic Use in ICU
Patients, 1995-2002

Resistance
Exposure
attributable
fraction
a,b

MDR
Organism
†

XDR
Organism
††

Any Resistant
Organism

Any antibiotic use 56.7% (30.5%, 73.0%) 81.2% (23.0%, 95.4%) 61.3% (39.5%, 75.3%)

Through the
kidneys
40.3% (1.2%, 63.9) 57.2% (-88.0%, 90.3%) 41.7% (6.1%, 63.8%)

Through the liver 68.6% (48.9%, 80.7%) 88.3% (51.6%, 97.2%) 72.8% (57.0%, 82.9%)

a
Odds ratios from multinomial logistic regression adjusted for age and ICU LOS using no
organism as the reference group for resistance and no antibiotics as the reference group for
elimination route were used in the AF calculation.
b
N=11,563 patients contributed to the “any antibiotic use”, 7,205 patients contributed to the
“through the kidneys”, and 6,396 patients contributed to the “through the liver” groups.
†
Multi-drug resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

113

Table 3.9. Fraction of Drug Resistance to ESKAPE Organisms Attributable to
Antibiotic Use in ICU Patients, 1995-2002

Resistance
Exposure attributable fraction
a,b
Any Resistant Organism

Any antibiotic use 52.2% (5.12%, 76.0%)

Through the kidneys 35.7% (-33.5%, 69.1%)

Through the liver 63.1% (24.9%, 81.9%)

a
Odds ratios from multinomial logistic regression adjusted for age and ICU LOS using no
organism as the reference group for resistance and no antibiotics as the reference group for
elimination route were used in the AF calculation.
b
N=10,896 patients contributed to the “any antibiotic use”, 6,952 patients contributed to the
“through the kidneys”, and 5,944 patients contributed to the “through the liver” groups.
†
Multi-drug resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

114

Figure 3.1. Patient Flowchart

Patients with
Medications
Patients with Positive
Blood Culture
Patients in
Hospital and ICU
N=14,288
Patients in
ICU only
N=430
Patients in ICU Cohort
N=14,718
Electronic IDs in ICU
with Antibiotics
N=12,849
Electronic IDs in ICU
with Medications
N=15,834
Exclude Electronic
IDs in ICU with
Antibiotics but not
in Hospital Data
N=1,874
Patients with Positive
Blood Cultures
N=12,627
Patients with Eligible
Blood Cultures
N=8,847
Patients in ICU
Cohort with Eligible
Blood Cultures
N=2,473
Patients in Analytic Cohort (1995-2002)
N=12,001
Exclude Patients with all Data Prior to 1995
N=2,717
Final Analytic Cohort (1995-2002)
N=11,576
Exclude Patients with Age <18 years
N=425
Patients in
Hospital and ICU
Patients with Eligible
Blood Culture
N=1,079
Patients with
Antibiotics
N=9,541
Patients in ICU Cohort
with Antibiotic
E_ID: Hosp ID:
N=10,975 N=9,894
Exclude Patients with
ineligible isolates
N=3,780
Ineligible isolates
N1=15

Through Kidneys
N=5,176
Through Liver
N=4,365
XDR
Organism
N=115
MDR
Organism
N=398
Non-resistant
Organism
N=566
Patients with No
Antibiotics
N=2,035
Patients with No
Organism
N=10,497

115

Figure 3.2. Distribution of antibiotics among ICU patients receiving antibiotics
eliminated through the kidneys or liver
Kidneys, only

Liver, any

116
Figure 3.3. Distribution of Select Antibiotics over Time among ICU Patients, 1995-2002

117

Figure 3.4. Distribution of gram negative and gram positive organisms among
ICU patients with HA-BSI organisms
Gram negative organisms

Gram positive organisms

118
Figure 3.5. Distribution of Select Organisms over Time among ICU Patients with a Positive Blood Culture, 1995-2002

119

Figure 3.6. Distribution of Organism Resistance among Patients with a Positive
Blood Culture, by Admitting Year

120

3.9 REFERENCES
1. Heron, M., Deaths: Leading Causes for 2014. Natl Vital Stat Rep, 2016.
65(5): p. 1-96.
2. Wisplinghoff, H., et al., Nosocomial bloodstream infections in US
hospitals: analysis of 24,179 cases from a prospective nationwide
surveillance study. Clin Infect Dis, 2004. 39(3): p. 309-17.
3. Infectious Disease Society of America, Bad Bugs, No Drugs: as antibiotic
discovery stagnates, a public health crisis brews, 2004.
4. Centres for Disease Control and Prevention, Antibiotic Resistance Threats
in the United States, 2013. Centres for Disease Control and Prevention,
US Department of Health and Human Services.
5. Boucher, H.W., et al., Bad bugs, no drugs: no ESKAPE! An update from
the Infectious Diseases Society of America. Clin Infect Dis, 2009. 48(1): p.
1-12.
6. World Health Organization, Antimicrobial Resistance: Global Report on
Surveillance, 2014. World Health Organization.
7. Dethlefsen, L., et al., The pervasive effects of an antibiotic on the human
gut microbiota, as revealed by deep 16S rRNA sequencing. PLoS Biol,
2008. 6(11): p. e280.
8. Jakobsson, H.E., et al., Short-term antibiotic treatment has differing long-
term impacts on the human throat and gut microbiome. PLoS One, 2010.
5(3): p. e9836.

121

9. Jernberg, C., et al., Long-term ecological impacts of antibiotic
administration on the human intestinal microbiota. ISME J, 2007. 1(1): p.
56-66.
10. Bodro, M., et al., Epidemiology, antibiotic therapy and outcomes of
bacteremia caused by drug-resistant ESKAPE pathogens in cancer
patients. Support Care Cancer, 2014. 22(3): p. 603-10.
11. Vazquez-Guillamet, M.C., et al., Predicting Resistance to Piperacillin-
Tazobactam, Cefepime and Meropenem in Septic Patients With
Bloodstream Infection Due to Gram-Negative Bacteria. Clin Infect Dis,
2017. 65(10): p. 1607-1614.
12. Dethlefsen, L. and D.A. Relman, Incomplete recovery and individualized
responses of the human distal gut microbiota to repeated antibiotic
perturbation. Proc Natl Acad Sci U S A, 2011. 108 Suppl 1: p. 4554-61.
13. Los Angeles County + University of Southern California Medical Center,
LAC+USC Medical Center Rules and Regulations, 2014.
14. Centers for Medicare and Medicaid Services. ICD-9-CM Diagnosis and
Procedure Codes: Abbreviated and Full Code Titles. 2014 [cited 2017
10Mar]; Available from:
https://www.cms.gov/Medicare/Coding/ICD9ProviderDiagnosticCodes/cod
es.html.
15. Truven Health Analytics, Micromedex.
16. Clinical and Laboratory Standards Institute, Performance Standards for
Antimicrobial Susceptibility Testing, CLSI supplement M100.

122

17. Magiorakos, A.P., et al., Multidrug-resistant, extensively drug-resistant and
pandrug-resistant bacteria: an international expert proposal for interim
standard definitions for acquired resistance. Clin Microbiol Infect, 2012.
18(3): p. 268-81.
18. Zhang, L., et al., Antibiotic administration routes significantly influence the
levels of antibiotic resistance in gut microbiota. Antimicrob Agents
Chemother, 2013. 57(8): p. 3659-66.
19. Chantziaras, I., et al., Studying the effect of administration route and
treatment dose on the selection of enrofloxacin resistance in commensal
Escherichia coli in broilers. J Antimicrob Chemother, 2017. 72(7): p. 1991-
2001.
20. Ghibellini, G., E.M. Leslie, and K.L. Brouwer, Methods to evaluate biliary
excretion of drugs in humans: an updated review. Mol Pharm, 2006. 3(3):
p. 198-211.
21. Karakullukcu, A., et al., Determination of clinical significance of coagulase-
negative staphylococci in blood cultures. Diagn Microbiol Infect Dis, 2017.
87(3): p. 291-294.
22. Hitzenbichler, F., et al., Clinical significance of coagulase-negative
staphylococci other than S. epidermidis blood stream isolates at a tertiary
care hospital. Infection, 2017. 45(2): p. 179-186.
23. Healy, C.M., et al., Distinguishing true coagulase-negative Staphylococcus
infections from contaminants in the neonatal intensive care unit. J
Perinatol, 2013. 33(1): p. 52-8.

123

24. Mehta, M.S., et al., Identification of Staphylococcus species directly from
positive blood culture broth by use of molecular and conventional
methods. J Clin Microbiol, 2009. 47(4): p. 1082-6.
25. Becker, K., C. Heilmann, and G. Peters, Coagulase-negative
staphylococci. Clin Microbiol Rev, 2014. 27(4): p. 870-926.
26. John, J., et al., Staphylococcus epidermidis and other Coagulase-
Negative Staphylococci. Antimicrobe: Infectious Disease Antimicrobial
Agents.

124

3.10 SUPPLEMENTARY MATERIAL
Supplementary Table 3.1. Admitting Diagnoses assessed for ICU Patients

Diagnosis Category Diagnosis
Infectious and Parasitic
Diseases Intestinal Infectious Diseases

Tuberculosis

Zoonotic Bacterial Diseases

Other Bacterial Diseases

Human Immunodeficiency Virus
Poliomyelitis And Other Non-Arthropod-Borne Viral Diseases Of
Central Nervous System

Viral Diseases Generally Accompanied By Exanthem

Arthropod-Borne Viral Diseases

Other Diseases Due To Viruses And Chlamydiae

Rickettsioses And Other Arthropod-Borne Diseases

Syphilis And Other Venereal Diseases

Other Spirochetal Diseases

Mycoses

Helminthiases

Other Infectious And Parasitic Diseases

Late Effects Of Infectious And Parasitic Diseases
Neoplasms
Malignant Neoplasm Of Lip, Oral Cavity, And Pharynx

Malignant Neoplasm Of Digestive Organs And Peritoneum

Malignant Neoplasm Of Respiratory And Intrathoracic Organs
Malignant Neoplasm Of Bone, Connective Tissue, Skin, And
Breast

Malignant Neoplasm Of Genitourinary Organs

Malignant Neoplasm Of Other And Unspecified Sites

Malignant Neoplasm Of Lymphatic And Hematopoietic Tissue

Neuroendocrine tumors

Benign Neoplasms

Carcinoma In Situ

Neoplasms Of Uncertain Behavior

Neoplasms Of Unspecified Nature
Endocrine, Nutritional and
Metabolic Diseases, and
Immunity Disorders Disorders Of Thyroid Gland

Diseases Of Other Endocrine Glands

Nutritional Deficiencies

Other Metabolic Disorders And Immunity Disorders
Diseases of the Blood and
Blood Forming Organs Iron deficiency anemias

Other deficiency anemias

Hereditary hemolytic anemias

Acquired hemolytic anemias

Aplastic anemia and other bone marrow failure syndromes

Other and unspecified anemias

Coagulation defects

125

Supplementary Table 3.1. continued
Diagnosis Category Diagnosis

Purpura and other hemorrhagic conditions

Diseases of white blood cells

Other diseases of blood and blood-forming organs
Mental Disorders
Organic Psychotic Conditions

Other Psychoses
Neurotic Disorders, Personality Disorders, And Other
Nonpsychotic Mental Disorders

Intellectual Disabilities
Diseases of the Nervous
System and Sense Organs Inflammatory Diseases Of The Central Nervous System
Hereditary And Degenerative Diseases Of The Central Nervous
System

Pain

Other Headache Syndromes

Other Disorders Of The Central Nervous System

Disorders Of The Peripheral Nervous System

Disorders Of The Eye And Adnexa

Diseases Of The Ear And Mastoid Process
Diseases of the Circulatory
System Acute Rheumatic Fever

Chronic Rheumatic Heart Disease

Hypertensive Disease

Ischemic Heart Disease

Diseases Of Pulmonary Circulation

Other Forms Of Heart Disease

Cerebrovascular Disease

Diseases Of Arteries, Arterioles, And Capillaries
Diseases Of Veins And Lymphatics, And Other Diseases Of
Circulatory System
Diseases of the Respiratory
System Acute Respiratory Infections

Other Diseases Of Upper Respiratory Tract

Pneumonia And Influenza

Chronic Obstructive Pulmonary Disease And Allied Conditions
Pneumoconioses And Other Lung Diseases Due To External
Agents

Other Diseases Of Respiratory System
Diseases of the Digestive
System Diseases Of Oral Cavity, Salivary Glands, And Jaws

Diseases Of Esophagus, Stomach, And Duodenum

Appendicitis

Hernia Of Abdominal Cavity

Noninfective Enteritis And Colitis

Other Diseases Of Intestines And Peritoneum

Other Diseases Of Digestive System
Diseases of the
Genitourinary System Nephritis, Nephrotic Syndrome, And Nephrosis

126

Supplementary Table 3.1. continued
Diagnosis Category Diagnosis

Other Diseases Of Urinary System

Diseases Of Male Genital Organs

Disorders Of Breast

Inflammatory Disease Of Female Pelvic Organs

Other Disorders Of Female Genital Tract
Complications of
Pregnancy, Childbirth, and
the puerperium
Ectopic And Molar Pregnancy And Other Pregnancy With
Abortive Outcome

Complications Mainly Related To Pregnancy
Normal Delivery, And Other Indications For Care In Pregnancy,
Labor, And Delivery
Complications Occurring Mainly In The Course Of Labor And
Delivery

Complications Of The Puerperium

Other Maternal And Fetal Complications
Diseases of the Skin and
Subcutaneous Tissue Infections Of Skin And Subcutaneous Tissue
Other Inflammatory Conditions Of Skin And Subcutaneous
Tissue

Other Diseases Of Skin And Subcutaneous Tissue

Diseases of the
Musculoskeletal System
and Connective Tissue Arthropathies And Related Disorders

Dorsopathies

Rheumatism, Excluding The Back
Osteopathies, Chondropathies, And Acquired Musculoskeletal
Deformities
Congenital Anomalies
Anencephalus and similar anomalies

Spina bifida

Other congenital anomalies of nervous system

Congenital anomalies of eye

Congenital anomalies of ear, face, and neck
Bulbus cordis anomalies and anomalies of cardiac septal
closure

Other congenital anomalies of heart

Other congenital anomalies of circulatory system

Congenital anomalies of respiratory system

Cleft palate and cleft lip

Other congenital anomalies of upper alimentary tract

Other congenital anomalies of digestive system

Congenital anomalies of genital organs

Congenital anomalies of urinary system

Certain congenital musculoskeletal deformities

Other congenital anomalies of limbs

Other congenital musculoskeletal anomalies

Congenital anomalies of the integument

Chromosomal anomalies

127

Supplementary Table 3.1. continued
Diagnosis Category Diagnosis

Other and unspecified congenital anomalies
Certain Conditions
Originating in the Perinatal
Period Maternal Causes Of Perinatal Morbidity And Mortality

Other Conditions Originating In The Perinatal Period
Symptoms, Signs, and Ill
Defined Conditions Symptoms

Nonspecific Abnormal Findings

Ill-Defined And Unknown Causes Of Morbidity And Mortality
Injury and Poisoning
Fracture Of Skull

Fracture Of Spine And Trunk

Fracture Of Upper Limb

Fracture Of Lower Limb

Dislocation

Sprains And Strains Of Joints And Adjacent Muscles

Intracranial Injury, Excluding Those With Skull Fracture

Internal Injury Of Chest, Abdomen, And Pelvis

Open Wound Of Head, Neck, And Trunk

Open Wound Of Upper Limb

Open Wound Of Lower Limb

Injury To Blood Vessels
Late Effects Of Injuries, Poisonings, Toxic Effects, And Other
External Causes

Superficial Injury

Contusion With Intact Skin Surface

Crushing Injury

Effects Of Foreign Body Entering Through Orifice

Burns

Injury To Nerves And Spinal Cord

Certain Traumatic Complications And Unspecified Injuries

Poisoning By Drugs, Medicinals And Biological Substances
Toxic Effects Of Substances Chiefly Nonmedicinal As To
Source

Other And Unspecified Effects Of External Causes
Complications Of Surgical And Medical Care, Not Elsewhere
Classified
Supplementary
Classification of External
Causes of Injury and
Poisoning External Cause Status
Activity
Railway Accidents
Motor Vehicle Traffic Accidents
Motor Vehicle Nontraffic Accidents
Other Road Vehicle Accidents
Water Transport Accidents
Air And Space Transport Accidents

128

Supplementary Table 3.1. continued
Diagnosis Category Diagnosis
Vehicle Accidents, Not Elsewhere Classifiable

Accidental Poisoning By Drugs, Medicinal Substances, And
Biologicals

Accidental Poisoning By Other Solid And Liquid Substances,
Gases, And Vapors
Misadventures To Patients During Surgical And Medical Care

Surgical And Medical Procedures As The Cause Of Abnormal
Reaction Of Patient Or Later Complication, Without Mention Of
Misadventure At The Time Of Procedure
Accidental Falls
Accidents Caused By Fire And Flames
Accidents Due To Natural And Environmental Factors

Accidents Caused By Submersion, Suffocation, And Foreign
Bodies
Other Accidents
Late Effects Of Accidental Injury

Drugs, Medicinal And Biological Substances Causing Adverse
Effects In Therapeutic Use
Suicide And Self-Inflicted Injury
Homicide And Injury Purposely Inflicted By Other Persons
Legal Intervention

Injury Undetermined Whether Accidentally Or Purposely
Inflicted
Injury Resulting From Operations Of War
Supplementary
Classification of Factors
Influencing Health Status
and Contact with Health
Services
Persons With Potential Health Hazards Related To
Communicable Diseases
Persons With Potential Health Hazards Related To Personal
And Family History
Persons Encountering Health Services In Circumstances
Related To Reproduction And Development

Liveborn Infants According To Type Of Birth

Persons With A Condition Influencing Their Health Status
Persons Encountering Health Services For Specific Procedures
And Aftercare

Persons Encountering Health Services In Other Circumstances
Persons Without Reported Diagnosis Encountered During
Examination And Investigation Of Individuals And Populations

Genetics

Body Mass Index

Estrogen Receptor Status
Other Specified Personal Exposures And History Presenting
Hazards To Health

Acquired Absence Of Other Organs And Tissue

Other Suspected Conditions Not Found

Retained Foreign Body

Multiple Gestation Placenta Status

129

Supplementary Table 3.2. Select Comorbidities assessed for ICU Patients

Comorbidity Category Comorbidity
Bone Fracture
Osteoporotic fracture
Osteoporosis
Cancer Cancer
Non-melanoma skin cancer
Cardiovascular Angina, stable
Angina, unstable
Heart failure
Hyperlipidemia
Hypertension
Myocardial infarction
Peripheral vascular disease
Stroke
Transient ischemic attack
Ventricular arrhythmia
Central Nervous System (CNS) Alzheimer’s
Dementia
Migraine
Parkinson’s disease
Connective Tissue Disorder Psoriatic arthritis
Psoriasis
Rheumatoid arthritis
Systemic lupus erythematosus
Endocrine Diabetes mellitus (Type 2)
Gastro-intestinal Crohn’s disease
GI Bleed
Irritable Bowel Syndrome
Ulcerative colitis
Hepatic Chronic liver disease
Cirrhosis
Pulmonary Asthma
Chronic obstructive pulmonary disease (CODP)
Renal Chronic kidney disease (CKD) without dialysis
Chronic kidney disease (CKD) with dialysis

Comorbidities are assessed using a 1 inpatient (IP) criteria in the baseline period.

130

Supplementary Table 3.3. Classification of Antibiotics Eliminated through the
Kidneys or through the Liver

Through the Kidneys At least Partially Through the Liver
Cefazolin Metronidazole
Piperacillin and Tazobactam Clindamycin
Vancomycin Erythromycin
Levofloxacin Oxacillin
Gentamicin Ceftriaxone
Cefotaxime Azithromycin
Ceftazidime Linezolid
Ampicillin and Sulbactam Clarithromycin
Imipenem Doxycycline
Ampicillin Chloramphenicol
Sulfamethoxazole and Trimethoprim Trovafloxacin
Ofloxacin Sulfadiazine
Tobramycin Quinupristin and Dalfopristin
Aztreonam Benzathine Penicillin
Cefepime Nitrofurantoin
Amikacin Tetracycline
Penicillin Polymyxin
Ciprofloxacin Atovaquone
Cefuroxime Moxifloxacin
Piperacillin Minocycline
Meropenem Paromomycin
Amoxicillin Alatrofloxacin
Cefotetan Dactinomycin
Cefalexin Mafenide
Cefoxitin Nafcillin
Streptomycin

Ertapenem

Amoxicillin Clavulanate

Imipenem and Cilastatin

Colistimethate

Netilmicin

Ceftizoxime

Cefapirin

Methenamine

Trimethoprim

Cefaclor

Cefixime

Ticarcillin and Clavulanate

Carbapenem

131

Supplementary Table 3.3. continued

Through the Kidneys At least Partially Through the Liver
Norfloxacin

Sulfamethoxazole

Tazobactam

Tobramycin and Vancomycin

132

Supplementary Table 3.4. Sensitivity Analysis for the Association between
Antibiotic Elimination Route and Drug Resistance of HA-BSI Organisms in ICU
Patients using Multinomial Model, 1995-2002

Resistance

Route
No
Organism
Non-resistant
Organism
MDR
Organism
†

XDR
Organism
††

No antibiotic use
N 1,915 63 19 2

Any antibiotic use
N 7,600 503 379 113
OR
a
(95% CI) 1.0 1.22 (0.93, 1.61) 2.52 (1.57, 4.04) 5.79 (1.42, 23.71)
p-value -- 0.1507 0.0001 0.0145

Through the
kidneys

N 4,448 225 113 24
OR
a
(95% CI) 1.0 1.21 (0.91, 1.62) 1.93 (1.18, 3.16) 3.66 (0.86, 15.52)
p-value -- 0.1856 0.0090 0.0785

Through the liver
N 3,152 278 266 89
OR
a
(95% CI) 1.0 1.29 (0.95, 1.73) 3.42 (2.11, 5.56) 8.70 (2.10, 35.96)
p-value -- 0.1015 <0.0001 0.0028

N=10,594 patients were included in the analysis; N=969 patients with ineligible organisms are
excluded from the baseline group of no organisms. Contrast to Table 4, in which patients with
ineligible organisms are classified as having no organisms in the “no organism” group.13
patients were excluded due to missing resistant scores.
a
Odds ratios are reported from multinomial logistic regression adjusted for age and ICU LOS
using no organism as the reference group for resistance and no antibiotics as the reference
group for elimination route.
†
Multi-drug resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

133

Supplementary Table 3.5. Sensitivity Analysis for the Association between
Antibiotic Elimination Route and Drug Resistance of HA-BSI Organisms in ICU
Patients using Cumulative Model, 1995-2002
No
Organism
Non-Resistant
Organism
MDR
Organism
†
XDR
Organism
††

No antibiotic use
N 1,915 63 19 2

Any antibiotic use
N 7,600 503 379 113
OR
a
(95% CI) 1.0 1.70 (1.35, 2.15) 3.10 (1.98, 4.85) 7.13 (1.74, 29.22)
p-value -- <0.0001 <0.0001 0.0063

Through the
kidneys

N 4,448 225 113 24
OR
a
(95% CI) 1.0 1.46 (1.14, 1.87) 2.21 (1.39, 3.52) 4.18 (0.99, 17.73)
p-value -- 0.0025 0.0008 0.0524

Through the liver
N 3,152 278 266 89
OR
a
(95% CI) 1.0 2.07 (1.62, 2.66) 4.41 (2.79, 6.97) 11.12 (2.68, 46.05)
p-value -- <0.0001 <0.0001 0.0009

N=10,594 patients were included in the analysis; N=969 patients with ineligible organisms are
excluded from the baseline group of no organisms. Contrast to Table 5, in which patients with
ineligible organisms are classified as having no organisms in the “no organism” group.13
patients were excluded due to missing resistant scores.
a
Odds ratios are reported from partial proportional odds model adjusted for age and duration in
the ICU. Resistance gradient is from less severe to more severe antibiotic resistance.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

134

Supplementary Table 3.6. Association between Type of Antibiotic as an Indicator
for Increasing Gut Exposure to Antibiotic Use and Drug Resistance of HA-BSI
Organisms in ICU Patients, 1995-2002

No
Organism
Non-
Resistant
Organism
MDR
Organism
†
XDR
Organism
††

Antibiotic use
N, no antibiotic use 1,951 63 19 2
N, through the kidneys 4,808 225 113 24
N, through the liver 3,725 278 266 89

Antibiotic use
a

Multinomial Model
b

OR (95% CI) 1.0 1.04 (0.90, 1.19) 1.69 (1.40, 2.40) 2.38 (1.58, 3.57)
p-value -- 0.6175 <0.0001 <0.0001
Cumulative Model
c

OR (95% CI) 1.0 1.33
‡
(1.19, 1.48) 1.90
‡
(1.60, 2.25) 2.69
‡
(1.79, 4.04)
p-value -- <0.0001 <0.0001 <0.0001

N=11,563 patients were included in the analysis; 13 patients were excluded due to missing
resistant scores.
a
Antibiotic use is modeled continuously, where 0=no antibiotic exposure, 1=through the
kidneys, 2=through the liver.
b
Odds ratios are reported from multinomial logistic regression model adjusted for age and
duration in the ICU using no organism as the reference group for resistance.
c
Odds ratios are reported from partial proportional odds model adjusted for age and duration in
the ICU. Resistance gradient is from less severe to more severe antibiotic resistance.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

135

Supplementary Table 3.7. Sensitivity Analysis for the Association between Type
of Antibiotic as an Indicator for Increasing Gut Exposure to Antibiotic Use and
Drug Resistance of HA-BSI Organisms in ICU Patients, 1995-2002

No
Organism
Non-
Resistant
Organism
MDR
Organism
†
XDR
Organism
††

Antibiotic use
c

N, no antibiotic use 1,915 63 19 2
N, through the kidneys 4,448 225 113 24
N, through the liver 3,152 278 266 89

Antibiotic Use
a

Multinomial Model
b

OR (95% CI) 1.0 1.11 (0.97, 1.28) 1.82 (1.51, 2.20) 2.56 (1.71, 3.85)
p-value -- 0.1270 <0.0001 <0.0001
Cumulative Model
c

OR (95% CI) 1.0 1.43
‡
(1.28, 1.60) 2.04
‡
(1.72, 2.42) 2.88
‡
(1.91, 4.33)
p-value -- <0.0001 <0.0001 <0.0001

N=969 patients with ineligible organisms are excluded from the baseline group of no
organisms. Contrast to Supplementary Table 6, in which patients with ineligible organisms are
classified as having no organisms in the “no organism” group.
N=10,594 patients were included in the analysis; 13 patients were excluded due to missing
resistant scores.
a
Antibiotic use is modeled continuously, where 0=no antibiotic exposure, 1=through the
kidneys, 2=through the liver.
b
Odds ratios are reported from multinomial logistic regression model adjusted for age and
duration in the ICU using no organism as the reference group for resistance.
c
Odds ratios are reported from partial proportional odds model adjusted for age and duration
in the ICU. Resistance gradient is from less severe to more severe antibiotic resistance.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

136

Supplementary Table 3.8. Sensitivity Analysis for the Association between
Antibiotic Elimination Route and Drug Resistance of HA-BSI Organisms Among
ICU Patients with Antibiotic Use, 1995-2002

Resistance
No
Organism
Non-resistant
Organism
Organism with
MDR
†
Organism with
XDR
††

Through the
kidneys

N 4,448 225 113 24

Through the liver

N 3,152 278 266 89
Multinomial
Model
a

OR (95% CI) 1.0 1.07 (0.88, 1.31) 1.79 (1.40, 2.28) 2.40 (1.48, 3.88)
p-value -- 0.5071 <0.0001 0.0004
Cumulative Model
b

OR (95% CI) 1.0 1.43 (1.23, 1.66) 2.00 (1.61, 2.49) 2.69 (1.66, 4.36)
p-value -- <0.0001 <0.0001 <0.0001

N=933 patients with ineligible organisms (360 patients with antibiotics eliminated through the
kidneys, and 573 patients with antibiotics eliminated through the liver) were excluded from the
baseline group of no organisms. Contrast to Table 6, in which patients with ineligible organisms
are classified as having no organisms in the “no organism” group.
N=8,595 patients were included in the analysis; 13 patients were excluded due to missing
resistant scores.
a
Odds ratios are reported from multinomial logistic regression model adjusted for age and
duration in the ICU using no organism as the reference group for resistance score.
b
Odds ratios are reported from partial proportional odds model adjusted for age and duration in
the ICU, and represent the odds of resistance of the observed level or greater, versus
everything less severe.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

137

Supplementary Table 3.9. Association between Antibiotic Elimination Route and
Drug Resistance of HA-BSI Organisms in ICU Patients, by Race, 1995-2002

Resistance

Route
No
organism
Non-resistant
Organism
MDR
Organism
†

XDR Organism
††

OR (95% CI) OR (95% CI) OR (95% CI)

No antibiotic use
Overall, N 1,951 63 19 2

Asian, N 200 3 2 0

Black, N 277 16 4 0

Hispanic, N 1,006 34 11 1

White, N 288 9 1 1

Any antibiotic
elimination

Overall, N 8,533 503 379 113
Odds Ref 1.12 (0.85, 1.48) 2.31 (1.44, 3.70) 5.32 (1.30, 21.76)
p-value -- 0.4136 0.0005 0.0201

Asian, N 777 42 35 10
Odds Ref 2.66 (0.80, 8.81) 2.82 (0.66, 12.08) N/A (NA, N/A)
p-value -- 0.1101 0.1630 0.9685

Black, N 1,154 85 66 24
Odds Ref 0.80 (0.45, 1.42) 2.37 (0.84, 6.68) N/A (N/A, N/A)
p-value -- 0.4475 0.1018 0.9511

Hispanic, N 4,813 269 206 51
Odds Ref 1.02 (0.70, 1.48) 1.95 (1.04, 3.63) 4.20 (0.57, 30.88)
p-value -- 0.9399 0.0362 0.1589

White, N 1,290 71 49 21
Odds Ref 1.10 (0.53, 2.30) 4.57 (0.62, 33.96) 1.44 (0.18, 11.49)
p-value -- 0.7930 0.1373 0.7335

Through the
kidneys

Overall, N 4,808 225 113 24
Odds Ref 1.15 (0.86, 1.53) 1.82 (1.11, 2.97) 3.46 (0.82, 14.68)
p-value -- 0.3510 0.0169 0.0923

Odds ratios are reported from multinomial logistic regression adjusted for age and duration in
the ICU using no organism as the reference group for resistance score and no antibiotics as
the reference group for route.
Stratified results are presented for Asian, Black, Hispanic, and White race and ethnicities.
Stratified results for American Indians were not presented due to small sample size. Overall
results presented are for the entire ICU cohort.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

138

Supplementary Table 3.9. continued
Resistance

Route
No
organism
Non-resistant
Organism
MDR
Organism
†

XDR Organism
††

OR (95% CI) OR (95% CI) OR (95% CI)

Race
Asian, N 467 25 9 1
Odds Ref 3.02 (0.90, 10.16) 1.64 (0.35, 7.67) N/A (N/A, N/A)
p-value -- 0.0744 0.5316 0.9582

Black, N 587 30 19 7
Odds Ref 0.72 (0.39, 1.36) 1.81 (0.61, 5.41) N/A (N/A, N/A)
p-value -- 0.3122 0.2850 0.9524

Hispanic, N 2,759 121 68 11
Odds Ref 1.04 (0.70, 1.53) 1.70 (0.89, 3.24) 2.87 (0.37, 22.28)
p-value -- 0.8590 0.1059 0.3140

White, N 696 29 11 4
Odds Ref 1.04 (0.48, 2.25) 2.99 (0.38, 23.44) 0.92 (0.10, 8.48)
p-value -- 0.9186 0.2969 0.9397

Through the liver
Overall, N 3,725 278 266 89
Odds Ref 1.12 (0.83, 1.51) 2.99 (1.84, 4.85) 7.60 (1.84, 31.41)
p-value -- 0.4625 <0.0001 0.0051

Asian, N 310 17 26 9
Odds Ref 2.08 (0.57, 7.64) 4.85 (1.10, 21.52) N/A (N/A, N/A)
p-value -- 0.2688 0.0376 0.9497

Black, N 567 55 47 17
Odds Ref 0.90 (0.49, 1.68) 3.00 (1.04, 8.67) N/A (N/A, N/A)
p-value -- 0.7490 0.0421 0.9501

Hispanic, N 2,054 148 138 40
Odds Ref 1.01 (0.67, 1.52) 2.31 (1.21, 4.40) 5.76 (0.77, 43.11)
p-value -- 0.9782 0.0109 0.0882

White, N 594 42 38 17
Odds Ref 1.21 (0.55, 2.66) 6.58 (0.87, 49.74) 2.07 (0.25, 17.20)
p-value -- 0.6325 0.0681 0.5017

Odds ratios are reported from multinomial logistic regression adjusted for age and duration in
the ICU using no organism as the reference group for resistance score and no antibiotics as
the reference group for route.
Stratified results are presented for Asian, Black, Hispanic, and White race and ethnicities.
Stratified results for American Indians were not presented due to small sample size. Overall
results presented are for the entire ICU cohort.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

139

Supplementary Table 3.10. Association between Antibiotic Elimination Route and
Drug Resistance of HA-BSI Organisms in ICU Patients, by Sex, 1995-2002

Resistance

Route
No
Organism
Non-resistant
Organism
MDR Organism
†
XDR Organism
††

OR (95%
CI)
OR (95% CI) OR (95% CI) OR (95% CI)

No antibiotic use
Overall, N 1,946 63 19 2

Male, N 1,314 49 14 2

Female, N 632 14 5 0

Any antibiotic
elimination

Overall, N 8,529 503 379 113
Odds Ref 1.12 (0.85, 1.48) 2.31 (1.44, 3.70) 5.32 (1.30, 21.76)
p-value -- 0.4136 0.0005 0.0201

Male, N 5,642 332 268 78
Odds Ref 0.93 (0.68, 1.28) 2.21 (1.27, 3.83) 3.64 (0.88, 15.05)
p-value -- 0.6712 0.0049 0.0741

Female, N 2,887 171 111 35
Odds Ref 1.77 (1.01, 3.10) 2.54 (1.02, 6.35) N/A (N/A, NA)
p-value -- 0.0478 0.0460 0.9507

Through the kidneys
Overall, N 4,805 225 113 24
Odds Ref 1.15 (0.86, 1.53) 1.82 (1.11, 2.97) 3.46 (0.82, 14.68)
p-value -- 0.3510 0.0169 0.0923

Male, N 3,141 140 78 17
Odds Ref 0.94 (0.67, 1.31) 1.76 (0.99, 3.13) 2.51 (0.58, 10.91)
p-value -- 0.7172 0.0538 0.2194

Female, N 1,664 85 35 7
Odds Ref 1.84 (1.04, 3.28) 2.00 (0.78, 5.14) N/A (N/A, N/A)
p-value -- 0.0377 0.1521 0.9530

Through the liver
Overall, N 3,724 278 266 89
Odds Ref 1.12 (0.83, 1.51) 2.99 (1.84, 4.85) 7.60 (1.84, 31.41)
p-value -- 0.4625 <0.0001 0.0051

Odds ratios are reported from multinomial logistic regression adjusted for age and duration in
the ICU using no organism as the reference group for resistance score and no antibiotics as
the reference group for route. Results for organism with no resistance is not shown in the table.
N=11,554 patients were included in the analysis; 13 patients were excluded due to missing
resistant scores and 9 additional patients were excluded due to missing sex. Overall results
presented are for the entire ICU cohort.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

140

Supplementary Table 3.10. continued

Resistance

Route
No
Organism
Non-resistant
Organism
MDR Organism
†
XDR Organism
††

OR (95%
CI)
OR (95% CI) OR (95% CI) OR (95% CI)

Male, N 2,501 192 190 61
Odds Ref 0.96 (0.68, 1.35) 2.81 (1.59, 4.96) 5.01 (1.19, 21.02)
p-value -- 0.7966 0.0004 0.0278

Female, N 1,223 86 76 28
Odds Ref 1.68 (0.92, 3.06) 3.35 (1.31, 8.56) N/A (N/A, N/A)
p-value -- 0.0930 0.0117 0.9487

Odds ratios are reported from multinomial logistic regression adjusted for age and duration in
the ICU using no organism as the reference group for resistance score and no antibiotics as
the reference group for route. Results for organism with no resistance is not shown in the table.
N=11,554 patients were included in the analysis; 13 patients were excluded due to missing
resistant scores and 9 additional patients were excluded due to missing sex. Overall results
presented are for the entire ICU cohort.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

141

Supplementary Table 3.11. Association between Antibiotic Elimination Route and
Drug Resistance of HA-BSI Organisms in ICU Patients with a Positive Blood
Culture, by Gram Stain Classification, 1995-2002

Resistance

Route
Non-resistant
Organism
MDR Organism
†
XDR Organism
††

OR (95% CI) OR (95% CI) OR (95% CI)

No antibiotic use
Overall, N 63 19 2

Gram Positive, N 43 18 2

Gram Negative, N 20 1 0

Any antibiotic elimination
Overall, N 503 379 113
Odds Ref 1.98 (1.15, 3.40) 4.41 (1.05, 18.49)
p-value -- 0.0138 0.0428

Gram Positive, N 284 310 64
Odds Ref 2.03 (1.13, 3.66) 3.03 (0.70, 13.09)
p-value -- 0.0181 0.1368

Gram Negative, N 219 69 49
Odds Ref 5.05 (0.66, 39.01) N/A (N/A, N/A)
p-value -- 0.1202 0.9757

Through the kidneys
Overall, N 225 113 24
Odds Ref 1.55 (0.88, 2.73) 2.88 (0.66, 12.53)
p-value -- 0.1271 0.1596

Gram Positive, N 117 85 9
Odds Ref 1.63 (0.88, 3.02) 1.46 (0.30, 7.05)
p-value -- 0.1244 0.6368

Gram Negative, N 108 28 15
Odds Ref 4.65 (0.59, 36.38) N/A (N/A, N/A)
p-value -- 0.1434 0.9450

Through the liver
Overall, N 278 266 89
Odds Ref 2.54 (1.45, 4.46) 6.22 (1.46, 26.44)
p-value -- 0.0011 0.0133

Odds ratios are reported from multinomial logistic regression adjusted for age and duration in
the ICU using no organism as the reference group for resistance score and no antibiotics as
the reference group for route. Results for organism with no resistance is not shown in the table.
N=1,079 patients with a positive blood culture were included in the analysis; 13 patients were
excluded due to missing resistant scores.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

142

Supplementary Table 3.11. continued
Resistance

Route
Non-resistant
Organism
MDR Organism
†
XDR Organism
††

OR (95% CI) OR (95% CI) OR (95% CI)

Gram Positive, N 167 225 55
Odds Ref 2.50 (1.36, 4.62) 4.53 (1.03, 19.86)
p-value -- 0.0033 0.0450

Gram Negative, N 111 41 34
Odds Ref 5.77 (0.73, 45.91) N/A (N/A, N/A)
p-value -- 0.0974 0.9429

Odds ratios are reported from multinomial logistic regression adjusted for age and duration in
the ICU using no organism as the reference group for resistance score and no antibiotics as
the reference group for route. Results for organism with no resistance is not shown in the table.
N=1,079 patients with a positive blood culture were included in the analysis; 13 patients were
excluded due to missing resistant scores.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

143

Supplementary Table 3.12. Association between Antibiotic Elimination Route and
Drug Resistance of HA-BSI Organisms among ICU Patients with Antibiotic Use
and Positive Blood Culture, 1995-2002

Resistance

Non-resistant
Organism
MDR
Organism
†
XDR
Organism
††

Overall
a

Through the kidneys
N 225 113 24
Through the liver

N 278 266 89
Multinomial Model
b

OR (95% CI) 1.0 1.64 (1.21, 2.21) 2.17 (1.30, 3.62)
p-value -- 0.0013 0.0029
Cumulative Model
c

OR (95% CI) 1.0 1.72 (1.31, 2.26) 1.72 (1.31, 2.26)
p-value -- <0.0001 <0.0001

Without Staphyloccus Species
d

Through the kidneys
N 208 63 16
Through the liver
N 243 118 45
Multinomial Model
b

OR (95% CI) 1.0 1.39 (0.95, 2.03) 1.69 (0.89, 3.20)
p-value -- 0.0926 0.1064
Cumulative Model
c

OR (95% CI) 1.0 1.44 (1.03, 2.03) 1.44 (1.03, 2.03)
p-value -- 0.0356 0.0356

Only Staphyloccus Species
e

Through the kidneys
N 17 50 8
Through the liver
N 35 148 44
Multinomial Model
b

OR (95% CI) 1.0 1.43 (0.71, 2.89) 2.26 (0.84, 6.14)
p-value -- 0.3233 0.1085
Cumulative Model
c

OR (95% CI) 1.0 1.56 (0.88, 2.75) 1.56 (0.88, 2.75)
p-value -- 0.1278 0.1278

a
N=995 patients with both Staphylococcus species and non-Staphylococcus species were
included in the analysis.
b
Odds ratios are reported from multinomial logistic regression model
adjusted for age and duration in the ICU using organism with no resistance as the reference
group for resistance score.
c
Odds ratios are reported from proportional odds logistic regression
model adjusted for age and duration in the ICU. At the observed level, they represents the
odds of observed resistance or more severe resistance, versus less severe antibiotic
resistance, and are for the following groups: a) MDR and XDR versus no resistance, and b)
XDR versus less resistance.
d
N=693 patients with non-Staphylococcus species were included
in the analysis.
e
N=302 patients with Staphylococcus species were included in the analysis.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

144

Supplementary Table 3.13. Sensitivity Analysis for the Fraction of Drug
Resistance Attributable to Antibiotic Use in ICU Patients, 1995-2002

Resistance
MDR
Organism
†

XDR
Organism
††

Any Resistant
Organism

Exposure
attributable
fraction

Any antibiotic
use
60.3% (36.3%, 75.2%) 82.7% (29.4%, 95.8%) 64.5% (44.5%, 77.4%)

Through the
kidneys
43.7% (6.9%, 66.0%) 59.6% (77.3%, 90.8%) 45.1% (11.5%, 95.9%)

Through the
liver
72.6% (55.5%, 83.2%) 89.8% (57.8%, 97.5%) 76.4% (62.6%, 85.1%)

N=969 patients with ineligible organisms were excluded from the baseline group of no
organisms that are used to derive the risk estimates used for the attributable fractions. Contrast
to Table 8, in which patients with ineligible organisms are classified as having no organisms in
the “no organism” group.
N=10,594 patients were included in the analysis; 13 patients were excluded due to missing
resistant scores.
Odds ratios from multinomial logistic regression adjusted for age and ICU LOS using no
organism as the reference group for resistance and no antibiotics as the reference group for
elimination route were used in the attributable fraction calculations.
†
Multi-drug resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.

145

Chapter 4: Route of antibiotic elimination and subsequent antibiotic
resistance

146

Route of antibiotic elimination and subsequent antibiotic resistance
Shivani Aggarwal
1
, Emi Minejima
2
, Victoria K. Cortessis
1
, Roberta McKean-
Cowdin
1
, Thomas Mack
1,3,4*
, Wendy Cozen
1,3,4*
, Robert Larsen
5

1
Department of Preventive Medicine, Keck School of Medicine, University of
Southern California, Los Angeles, California
2
Department of Clinical Pharmacy, University of Southern California School of
Pharmacy, Los Angeles, California
3
Norris Comprehensive Cancer Center, Keck School of Medicine, University of
Southern California, Los Angeles, California

4
Department of Pathology, Keck School of Medicine, University of Southern
California, Los Angeles, California
5
Division of Infectious Diseases, Department of Medicine, Keck School of
Medicine, University of Southern California, Los Angeles, California

147

4.1 ABSTRACT
To interrogate the association between route of antibiotic elimination (liver
vs kidney) and drug resistance and mortality, we conducted an extreme
phenotype case-case study nested within a retrospective cohort study of ICU
patients hospitalized at the Los Angeles County + University of Southern
California Medical Center between 1995 and 2002 (N=11,563). Patients with a
qualifying bloodstream infection that were treated with at least one antibiotic
(N=693) were classified according to the route of antibiotic elimination and
according to drug resistance (none, multi-drug, or extensive-drug resistance) of
bacteria cultured from the bloodstream infection. Multinomial regression was
used to assess the association between the route of elimination and resistance.
Cox proportional hazards model was used to assess the association between
route of elimination and mortality. Patients with any antibiotic eliminated through
the liver had increased odds of a subsequent drug-resistant BSI of any kind
(OR=1.44, p=0.0390). The risk of death for patients with antibiotics eliminated
through the liver was 1.86 times that for patients with antibiotics eliminated
through the kidneys (p=0.007).
4.2 INTRODUCTION
Hospital-acquired infections are a major public health concern and are
associated with high morbidity and mortality. Hospital-acquired bloodstream
infections (HA-BSIs) are among the leading causes of death in the United States
(US), and an estimated 535,920 to 628,320 incident cases are reported annually
[1-3]. Patients with HA-BSIs have higher mortality and an excess of 8-14 days in

148

the hospital, compared to patients without [4-6]. Approximately half of all HA-
BSIs occur in Intensive Care Unit (ICU) patients, and 26-48% of these result in
death [7]. Rising occurrence of multi-drug resistance among patients with HA-
BSIs is a growing concern. Predictors of HA-BSI include increased length of stay
(LOS) in the hospital, catheter use, and mechanical ventilation [8-10]. The
ESKAPE bacteria Enterococcus faecium, Staphylococcus aureus, Klebsiella
pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and
Enterobacter species are the most common antibiotic resistant infectious agents
identified in the hospital setting [11, 12]. Prior antibiotic use is a risk factor for
subsequent organism resistance [8, 9], with risk ranging from 1.54-2.13, for
hospitalized patients with positive blood cultures within 30 days after antibiotic
use [9].
In vivo studies have demonstrated that gut bacteria can develop
resistance to antibiotics in as little as one day after exposure, presumably due to
selective pressure and horizontal transfer of resistant genes [13]. Microbiology
studies in humans and animals show that prior antibiotic use depletes
commensal bacteria diversity, thereby increasing the proportion of resistant
bacteria [8, 9, 13-16]. Degree and duration of exposure of gastrointestinal
organisms to antibiotics are likely to be influenced by the route through which the
antibiotics and their metabolites are eliminated, through either the liver and biliary
system versus through the kidneys. Antibiotics eliminated though the liver and
biliary system pass through the gastrointestinal tract and expose gut bacteria,
while antibiotics eliminated predominantly through the kidneys result in minimal

149

gastrointestinal exposure and less contact with bacteria. However, no
epidemiologic studies have directly assessed impact of the route of drug
elimination on antibiotic resistance.
We examined this relationship using existing hospital records of antibiotic
usage and the resistance patterns of organisms subsequently identified from
blood cultures among patients treated at the Los Angeles County University of
Southern California Medical Center (LAC-USC MC) ICU. We hypothesized that
patients treated with antibiotics eliminated through the liver and biliary tract are at
greater risk of developing subsequent drug resistant HA-BSIs, compared to
patients treated with antibiotics eliminated through the kidneys.
4.3 MATERIALS AND METHODS
Study Design and Subjects
The goal of the study was to assess associations between the route of
antibiotic elimination and subsequent drug resistance and mortality. Patients with
a positive blood culture were selected retrospectively from the medical and
surgical ICUs at the LAC-USC MC, a county teaching hospital, (Figure 1)
between 1993 and 2003. From the full set of patients with a positive blood
culture who had been treated with antibiotics, we excluded patients under 18
years of age and those not treated with antibiotics preceding the HA-BSI blood
culture. The remaining patients were classified according to degree of antibiotic
resistance of the cultured organism. The study was based on retrospective
record review without patient contact. Thus, informed consent was waived and

150

Institutional Review Board approval was granted from the University of Southern
California Health Sciences Institutional Review Board.
Data Collection
Collected information included inpatient admission and discharge date and
time, admitting diagnosis, LOS in the ICU (defined as the number of days in the
ICU); date of birth, sex, race/ethnicity, surgeries and procedures, discharge
disposition, discharge date, subsequent readmissions, and death. Information on
prescribed medications (medication name, administration date and time, dose,
and route of administration), mechanical ventilation, catheter use, blood
transfusions, and laboratory results by organ site was also collected.
Blood Cultures
Blood cultures were ordered for patients exhibiting clinical signs and
symptoms suggesting infection, collected at the onset of suspicion, and repeated
until negative, at the discretion of the treating physician. All blood cultures were
processed within the LAC-USC MC Microbiology Laboratory according to
standard procedures [17]. Positive blood cultures were reported through the
electronic medical record (EMR) and susceptibility testing was performed using a
standard panel of antimicrobial agents based on gram stain result.
Classification of Antibiotic Elimination Route
Route of elimination was classified first for each antibiotic and then
summarized for each patient. A minimum latency period of two days between
antibiotic exposure and development of the bacterial resistance profile was
assumed. The elimination route was classified based on whether adjustment for

151

renal function using creatinine clearance was required. If adjustment was
necessary, the antibiotic was presumed to be eliminated predominately through
the kidneys; otherwise, the antibiotic was assumed to be partially or
predominantly eliminated through the liver (Supplemental Table 1). The
cumulative effect of antibiotic elimination was thereby classified dichotomously as
either predominantly through the kidneys or at least partially through the liver.
Organism Resistance
Organisms isolated by blood cultures were identified. All non-bacterial
organisms (viruses, fungi, parasites) were excluded, and bacterial agents were
removed if either ill-defined or a likely to have resulted from contamination
(Bacillus cereus, Corynebacterium group G1, Corynebacterium group G2,
Corynebacterium jeikeium, Corynebacterium minutissimum, Corynebacterium
xerosis, unspecified Corynebacterium species, gram negative Coccobacillus,
coagulase negative staphylococcus, Staphylococcus epidermidis, and
Staphylococcus sciuri). Other organisms not commonly specified were also
removed (Bacillus species, unspecified Gram negative or positive rods, and
unspecified Staphylococcus species). Finally, bacteria for which reported gram
stain result was discordant with the cultured organism were removed.
Resistance to each antimicrobial agent was determined using standard
microbiology methods per CLSI and FDA recommendations [18]. Resistance
scores were derived using methods specified by the CDC and ECDC [19].
Microorganisms were categorized into Staphylococcus aureus, Enterococcus
species, Enterobacteriaceae family, Pseudomonas aeruginosa, or Acinetobacter

152

species. Gram negative isolates were classified as Enterobacteriaceae family,
Pseudomonas aeruginosa, or Acinetobacter species based on their resistance
profile, or classified as non-specific gram negative. Gram positive isolates were
classified as Staphylococcus aureus or Enterococcus species, or classified as
non-specific gram positive.
The resistance pattern was based on susceptibility data to the antibiotic
classes aminoglycoside, rifamycin, carbapenems, cephalosporins,
fluoroquinolones, folate pathway inhibitors, glycopeptides, lincosamides,
macrolides, monobactams, nitrofuran, nitroimidazoles, oxazolidinones,
penicillins, phenicols, polymyxins, streptogramins, sulfonamides, and
tetracyclines, and varied from organism to organism. Antibiotic resistance was
categorized as either no or minor resistance, multi-drug resistance (MDR) to at
least one antimicrobial agent in at least three antimicrobial classes, extensive-
drug resistance (XDR) to at least one antimicrobial agent all but two or less
antimicrobial classes, or pan-drug resistance (PDR) to antimicrobial agents in all
classes. Bacteria with intrinsic resistance to an antibiotic or antimicrobial class
did not contribute to the resistance derivation due to the potential for bias in the
drug resistance classification. For patients with a polymicrobial culture, the most
resistant organism was used for the analysis. Among patients with multiple blood
cultures over time, the first hospital-acquired blood culture that met the latency
requirement was selected.
Statistical Analysis

153

We summarized data using counts and proportions for categorical
variables; and means, standard deviations, and ranges for continuous variables.
Multinomial logistic regression was used to assess the strength of the association
between elimination route and organism resistance, where resistance was
classified as no resistance, MDR, or XDR. Cumulative logistic regression was
used to model organism resistance on an ordinal scale (at least MDR versus
less; XDR versus less), and the Proportional Odds assumption was tested. In
cases where the Proportional Odds assumption was not confirmed, partial
proportional odds logistics regression was used. We conducted sensitivity
analyses of the route-resistance association for variation in sex, gram stain
classification, age (<50 or ≥50), type of ICU (Medical or Surgical), and
race/ethnicity (Black or Hispanic). Finally, the Cox proportional hazards model
was used to assess the association between elimination route and mortality, with
patients followed up to the time of death in the hospital or discharge. We
assessed confounding by age, sex, race, ICU category (Medical or Surgical),
admission year, and number of days in the ICU by adding covariates individually
into the model, evaluating resulting impact on the beta coefficient of the main
effect. Covariates were included in final analytic models if their addition resulted
in an at least 15% change of magnitude of main effect beta estimates. Effect
modification by age, sex, race, and ICU category was assessed for all analyses.
All statistical analyses were performed using SAS software, Version 9.4 [Cary,
NC].

154

4.4 RESULTS
A total of 693 patients that met the criteria of prior antibiotic use and HA-
BSI were included in the analysis (Figure 1). The average (SD) age was 47.6
(16.6) years (Table 1). Approximately 66% of patients were male, 57.0%
Hispanic, 18.0% Black, and 13.7% non-Hispanic White. No resistance was
detected in organisms cultured from samples of 451 (65.1%) patients, while 191
(27.6%) had MDR-BSI, and 61 (8.8%) XDR HA-BSI. No patient had PDR HA-
BSI. The mean (SD) ICU LOS patients was 20.3 (21.3) days, and varied by
degree of antibiotic resistance. Patients with non-resistant, MDR, and XDR HA-
BSI had mean ICU LOS of 17.5 (18.0), 23.4 (24.0), and 31.6 (29.2) days,
respectively. The time from the start of treatment to blood culture was not
significantly different by route of antibiotic elimination, 147 (SD: 299) days for
patients treated with antibiotics eliminated through the kidneys and 105 (SD: 272)
days for those with antibiotics eliminated at least partially through the liver
(p>0.05) (Supplemental Figure 1).
The most commonly prescribed antibiotics eliminated predominantly
through the kidneys were Vancomycin (18.1%), Cefazolin (15.2%), Piperacillin
and Tazobactam (14.9%), Levofloxacin (11.9%), Cefotaxime (9.6%), and
Gentamicin (9.5%), and among those eliminated at least partially through the
liver were Metronidazole (34.1%), Clindamycin (17.2%), Erythromycin (13.6%),
Oxacillin (12.2%), and Azithromycin (7.7%). Gram positive organisms were more
frequently identified in culture (51.4%) than gram negative organisms (48.6%).
Staphylococcus aureus (43.3%) and Enterococcus species (30.6%) comprised

155

the majority of the gram positive organisms, while the most common gram-
negative organisms were Escherichia coli (19.9%), Klebsiella pneumoniae
(13.7%), Enterobacter cloacae (13.4%), Pseudomonas aeruginosa (12.5%), and
Acinetobacter baumannii (8.9%).
A total of 34.9% of patients had a culture with a resistant organism, of
which 26.1% were MDR and 8.8% were XDR. The most common MDR
organisms were Staphylococcus aureus (38.7%), Escherichia coli (11.6%),
Enterococcus species (9.9%), Enterococcus faecium (8.3%), and Acinetobacter
baumannii (7.7%). The most common XDR organisms were Pseudomonas
aeruginosa (14.8%), Klebsiella pneumoniae (14.8%), Enterobacter cloacae
(13.1%), Acinetobacter baumannii (13.1%), Staphylococcus aureus (6.6%),
Enterococcus faecium (6.6%), and Escherichia coli (4.9%) (Figure 2). Patients
with MDR HA-BSI had a higher (61.9%), while those with XDR HA-BSI had a
lower (19.7%), proportion of gram-positive bacteria compared to gram-negative
bacteria (Table 1).
Patients with prior use of antibiotics eliminated at least partially through
the liver had an increased risk of having a resistant HA-BSI (OR: 1.44, 95% CI:
1.02-2.04), and were 1.39 (95% CI: 0.95-2.03) and 1.69 (95% CI: 0.89-3.20)
times as likely to have an MDR and XDR HA-BSI, respectively (Table 2). No
significant difference in strength of the association between elimination route and
drug resistance were found between strata defined by gram stain classification,
median age (50 years), gender, type of ICU, or race (Supplemental Table 2).

156

By modeling resistance on an ordinal scale and assuming a standard change in
effect for each level increase in antibiotic use, we again found use of antibiotics
eliminated at least partially through the liver to be associated with greater
antibiotic resistance. Patients with prior use of antibiotics eliminated at least
partially through the liver were 1.44 (p=0.0356) times as likely to develop at least
MDR, with the same risk observed for XDR, compared to elimination
predominantly through the kidneys (data not shown).
Sensitivity analyses were performed limiting the analysis to patients with
blood cultures within 30 days, 31 days to 6 months, and greater than 6 months
from the start of antibiotic use (Table 3). Drug resistance was 57% more common
(OR: 1.57 [95% CI: 1.02-2.41], p=0.0405) for blood cultures within 30 days of the
antibiotic start date, and 62% (p=0.0111) for those between 31 days and 6
months of starting. Moreover, within the 30 day window use of antibiotics
eliminated through the liver was associated with significantly increased risk of
XDR (OR: 2.68 [95% CI: 1.15-6.25], p=0.0230), while patients with blood cultures
between 31 days and 6 months of the antibiotics start date had a significantly
increased risk of both MDR (OR: 1.53 [95% CI: 1.01-2.31], p=0.0049) and XDR
(OR: 1.95 [0.98-3.88], p=0.0563). However, among patients who developed
antibiotic-resistant blood cultures more than 6 months after the start of
antibiotics, resistance was not associated with antibiotics eliminated through the
liver.
Patients who used any antibiotics eliminated at least partially through the
liver experienced poorer survival compared to those who used antibiotics

157

eliminated through the kidneys (Figure 3). The age- and ICU LOS-adjusted risk
of death for patients with antibiotics eliminated at least partially through the liver
was 1.86 times that for elimination through the kidneys (p=0.007) (data not
shown).
4.5 DISCUSSION
Gut bacteria readily exchange antibiotic resistance via horizontal gene
transfer of mobile elements and plasmids encoding resistant genes, and that
antibiotic exposure to the gut can result in rapid and persistent antibiotic
resistance [13, 14, 16, 20, 21]. We found that ICU patients treated with antibiotics
eliminated at least partially through the liver had greater subsequent occurrence
of both MDR and XDR drug resistant HA-BSIs than those treated by antibiotics
eliminated through the kidneys. Moreover, use of antibiotics through the liver was
more strongly associated with occurrence of XDR resistance than with MDR
resistance. Finally, as expected if elimination route acutely impacts HA-BSI
resistance and has a less long-term impact on the gut, associations between
route and resistance appeared to be limited to organisms cultured from HA-BSIs
within 6 months of initiation of antibiotic treatment, with no apparent association
with resistance detected thereafter.
Other studies have reported varying proportions of gram negative ICU
infections, ranging from 58.3% in the EUROBACT international multi-cohort
prospective study [10] to only 25% in an 8 year prospective study of 49 US
hospitals [7]. In our study, proportion of patients who developed MDR and XDR
(26.1% and 8.8%, respectively) were lower than those in the EUROBACT study

158

(47.8% and 20.5%, respectively) [10], possibly because studied patients who
were treated during a period that preceded the EUROBACT study, when drug
resistance was less prevalent. All six of the ESKAPE organisms were identified in
our study population and represented 58% of all organisms in the primary
analysis. We found the distribution of the more common HA-BSI organisms
Staphylococcus aureus, Pseudomonas aeruginosa, and Escherichia coli to be
similar to that observed in other studies, although we found a lower percentage
of Acinetobacter baumannii and a higher percentage of Enterobacter species in
our study [7, 10].
A few studies have examined the relationship between pharmacologic
properties of antibiotics and resistance in animal models [22, 23]. The antibiotic
administration route was found to be associated with both the antibiotic
resistance gene pool and antibiotic resistant bacteria in the feces of mice
inoculated with non-resistant Enterococcus species or Escherichia coli [22].
Orally administered ampicillin conferred greater resistance compared to
intravenously (IV) administered ampicillin, while both orally and IV administered
tetracycline conferred greater resistance. This difference was likely due to the
difference in IV antibiotic exposure in the gastrointestinal tract: IV ampicillin is
excreted primarily through the kidneys whereas IV tetracycline is excreted both
through the liver and kidneys. Although epidemiologic studies have identified
prior antibiotic use as a risk factor of emergence of organism resistance, they
have not investigated antibiotic elimination route with respect of resistance in
humans with HA-BSIs [8, 9].

159

Strengths of our study include use of a clearly defined exposure based on
objective observations, and study of a large public hospital-based experience
spanning 8 years, making results generalizable to ICU patients of hospitals in the
US with similar patterns of antibiotic use and disease severity. A limitation is the
potential for confounding by indication whereby the severity of an infection might
both predispose to development of resistance and require antibiotics
preferentially eliminated through the liver or kidneys. However, the proportion of
patients diagnosed with a BSI during their ICU stay did not differ by elimination
route (p=0.4271), indicating that such confounding is unlikely. Another limitation
is the use of a categorical classification of gut exposure which broadly divides
elimination into “through kidneys” or “through liver” categories, rather than as a
continuous exposure that might more precisely capture any partial elimination of
some antibiotics through both routes. Although pharmacokinetic data on
antibiotics can be used to determine percent renal or hepatic elimination,
guidance is often unavailable or incomplete for less-common drugs. Moreover,
excretion into the bile and feces are more difficult to measure and may be
underreported [24], compared to excretion through urine. Thus, classifying
elimination on a continuous scale presents numerous challenges. Nonetheless,
any misclassification of the exposure definition would be expected to be non-
differential, and thus extremely unlikely to explain the strong association
observed between elimination route and resistance. A third limitation is the
inability to study influences of individual antibiotic preparations, because many
patients received multiple antibiotics.

160

We conclude that the route of antibiotic elimination can impact the risk for
development of drug resistance. These results, if reproduced, may influence
antibiotic selection and future antibiotic drug development in favor of compounds
eliminated through the renal route.

161

4.6 TABLES AND FIGURES
Table 4.1. Baseline and Demographic Characteristics of ICU Patients at Los
Angeles County Hospital (1995-2002) who were Included in the Study

All Patients
(N=693)
Non-resistant
BSI
(N=451)
MDR BSI
(N=181)
XDR BSI
(N=61)

Patient Characteristics
Age (years) N 693 451 181 61
Mean (Std) 47.6 (16.6) 47.1 (16.3) 49.1 (16.8) 46.8 (18.8)
Median 47.0 47.0 49.0 44.0
25th, 75th
percentile
35.0, 59.0 35.0, 58.0 36.0, 62.0 30.0, 62.0

Sex
Male n (%) 456 (65.8%) 296 (65.6%) 120 (66.3%) 40 (65.6%)
Female n (%) 237 (34.2%) 155 (34.4%) 61 (33.7%) 21 (34.4%)

Race/Ethnicity
Hispanic n (%) 361 (53.2%) 249 (56.3%) 88 (49.4%) 24 (41.4%)
Black n (%) 122 (18%) 70 (15.8%) 35 (19.7%) 17 (29.3%)
Asian n (%) 66 (9.7%) 39 (8.8%) 20 (11.2%) 7 (12.1%)
White n (%) 93 (13.7%) 61 (13.8%) 25 (14%) 7 (12.1%)
American Indian n (%) 2 (0.3%) 1 (0.2%) 1 (0.6%) 0 (0.0%)
Other n (%) 14 (2.1%) 8 (1.8%) 4 (2.2%) 2 (3.4%)
Unknown n (%) 20 (2.9%) 14 (3.2%) 5 (2.8%) 1 (1.7%)

Hospital Characteristics
ICU Category
SICU n (%) 349 (50.4%) 237 (52.5%) 87 (48.3%) 25 (41%)
MICU n (%) 307 (44.4%) 186 (41.2%) 88 (48.9%) 33 (54.1%)
NSICU n (%) 36 (5.2%) 28 (6.2%) 5 (2.8%) 3 (4.9%)

Length of Stay in
Hospital (days)
N 681 443 179 59
Mean (Std) 64.3 (94.7) 56.9 (99.7) 76.7 (85.7) 82.8 (75.0)
Median 37.0 32.0 48.0 55.0
25th, 75th
percentile
22.0, 68.0 20.0, 53.0 27.0, 95.0 34.0, 106.0

Length of Stay in ICU
(days)
N 693 451 181 61
Mean (Std) 20.3 (21.3) 17.5 (18.0) 23.4 (24.0) 31.6 (29.2)
Median 14.0 12.0 17.0 24.0
25th, 75th
percentile
6.0, 27.0 5.0, 24.0 6.0, 31.0 10.0, 41.0
SICU=Surgical ICU, MICU=Medical ICU, NICU=Neurosurgical ICU
Two patients with unknown and other sex were classified as having missing sex.
Four patients with a positive blood culture had an organism with unclassifiable gram stain.

162

Table 4.1. continued

All Patients
(N=693)
Non-resistant
BSI
(N=451)
MDR BSI
(N=181)
XDR BSI
(N=61)

Length of Stay in
Surgical ICU (days)
N 349 237 87 25
Mean (Std) 23.9 (23.9) 20.1 (18.7) 28.9 (28.6) 42.2 (36.1)
Median 17.0 14.0 20.0 38.0
25th, 75th
percentile
7.0, 31.0 7.0, 27.0 6.0, 37.0 12.0, 61.0

Length of Stay in
Medical ICU (days)
N 307 186 88 33
Mean (Std) 16.8 (18.3) 14.7 (17.6) 18.6 (17.9) 23.7 (21.6)
Median 11.0 9.0 14.0 16.0
25th, 75th
percentile
5.0, 21.0 4.0, 18.0 5.0, 27.0 9.0, 33.0

Length of Stay in
Neurosurgical ICU
(days)
N 36 28 5 3
Mean (Std) 15.5 (10.6) 14.1 (10.5) 14.6 (8.0) 29.3 (6.7)
Median 11.5 10.0 15.0 26.0
25th, 75th
percentile
9.0, 22.0 8.5, 17.0 11.0, 22.0 25.0, 37.0

Microbiological Characteristics
Gram positive n (%) 356 (51.4%) 232 (51.4%) 112 (61.9%) 12 (19.7%)
Gram negative n (%) 337 (48.6%) 219 (48.6%) 69 (38.1%) 49 (80.3%)

Gram positive
Staphylococcus
aureus
n (%) 154 (43.3%) 80 (34.5%) 70 (62.5%) 4 (33.3%)
Enterococcus
species
n (%) 109 (30.6%) 89 (38.4%) 18 (16.1%) 2 (16.7%)
Enterococcus
faecium
n (%) 22 (6.2%) 0 (0.0%) 0 (0.0%) 4 (33.3%)
Streptococcus,
Group A
n (%) 22 (6.2%) 17 (7.3%) 0 (0.0%) 0 (0.0%)
Enterococcus
faecalis
n (%) 12 (3.4%) 10 (4.3%) 0 (0.0%) 0 (0.0%)

Gram negative
Escherichia coli n (%) 67 (19.9%) 44 (19.8%) 21 (30.0%) 3 (6.1%)
Klebsiella
pneumoniae
n (%) 46 (13.7%) 31 (14.0 %) 6 (8.6%) 9 (18.4%)
Enterobacter
cloacae
n (%) 45 (13.4%) 31 (14.0%) 7 (10.0%) 8 (16.3%)

SICU=Surgical ICU, MICU=Medical ICU, NICU=Neurosurgical ICU
Two patients with unknown and other sex were classified as having missing sex.
Four patients with a positive blood culture had an organism with unclassifiable gram stain.

163

Table 4.1. continued

All Patients
(N=693)
Non-resistant
BSI
(N=451)
MDR BSI
(N=181)
XDR BSI
(N=61)

Pseudomonas
aeruginosa
n (%) 42 (12.46%) 34 (15.3%) 0 (0.0%) 9 (18.4%)
Acinetobacter
baumannii
n (%) 30 (8.9%) 9 (4.1%) 14 (20.0%) 8 (16.3%)
Enterobacter
aerogenes
n (%) 24 (7.1%) 18 (8.1%) 4 (5.7%) 2 (4.1%)

SICU=Surgical ICU, MICU=Medical ICU, NICU=Neurosurgical ICU
Two patients with unknown and other sex were classified as having missing sex.
Four patients with a positive blood culture had an organism with unclassifiable gram stain.

164

Table 4.2. Association between Antibiotic Elimination Route and Degree of
Resistant Bacteremia in Participating patients, 1995-2002

Resistance Score

Route

Organism with
no resistance
(N=451)
Organism with
MDR
resistance
†
(N=181)
Organism with
XDR
resistance
††
(N=61)
Organism with
Any
resistance
†††

(N=242)
OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI)

Through the
kidneys
Ref Ref Ref Ref

Through the
liver
1.0 (Ref) 1.39 (0.95, 2.03) 1.69 (0.89, 3.20) 1.44 (1.02, 2.04)
p-value -- 0.0926 0.1064 0.0390

Odds ratios are reported from multinomial logistic regression adjusted for age and duration in
the ICU using organism with no resistance as the reference group for resistance score.
N=693 patients were included in the analysis.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.
†††
Any resistance is defined as MDR or XDR resistance.

165

Table 4.3. Sensitivity Analysis of the Association between Antibiotic Elimination
Route and Degree of Resistant Bacteremia in Participating Patients, by Time
from Antibiotic Start Date, 1995-2002

Resistance Score

Window
a
Route
Organism
with no
resistance
Organism with
MDR
.resistance
†

Organism with
XDR
resistance
††

Organism with
Any
resistance
†††

OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI)

<= 30 days N=295 N=106 N=36 N=142
Through the
kidneys
Ref Ref Ref Ref

Through the
liver
1.0 (Ref) 1.27 (0.79, 2.05) 2.68 (1.15, 6.25) 1.50 (0.97, 2.32)
p-value -- 0.3185 0.0230 0.0686

Cumulative 1.0 (Ref) 1.57 (1.02, 2.41) N/A
p-value -- 0.0405

31 days to <= 6
months
N=372 N=158 N=55 N=213
Through the
kidneys
Ref Ref Ref Ref

Through the
liver
1.0 (Ref) 1.53 (1.01, 2.31) 1.95 (0.98, 3.88) 1.61 (1.11, 2.36)
p-value -- 0.0449 0.0563 0.0133

Cumulative 1.0 (Ref) 1.62 (1.12, 2.35) N/A
p-value -- 0.0111

> 6 months N=79 N=23 N=6 N=29
Through the
kidneys
Ref Ref Ref Ref

Through the
liver
1.0 (Ref) 0.64 (0.22, 1.84) 0.31 (0.03, 2.85) 0.55 (0.20, 1.52)
p-value -- 0.4047 0.2995 0.2503

Cumulative 1.0 (Ref) 0.52 (0.19, 1.40) N/A
p-value -- 0.1969

Odds ratios are reported from multinomial logistic regression adjusted for age of patient and
duration of stay in the ICU, using organism with no resistance as the reference group for
resistance score.
a
Window defined as time period from antibiotic start date to blood culture.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.
†††
Any resistance is defined as MDR or XDR resistance.

166

Figure 4.1. Patient Flowchart

Patients in Analytic
Cohort (1995-2002)
N=12,001
Final Analytic
Cohort (1995-2002)
N=11,576
Patients with Age
<18 years
N=425
Patients with No
Organism
N=10,497
Patients with Eligible Blood
Culture
N=1,079
Patients with Positive Blood
Culture and Antibiotics
N=995
Patients with No
Antibiotics
N=84
Non-resistant
Organism
Resistant Organism
N=242
Patients with Qualifying
Blood Culture and Antibiotics
N=693
Patients with
Staphylococcus species
N=302

167

Figure 4.2. Distribution of Organisms by Resistance Score
MDR, (N=181)

XDR, (N=61)

168

Figure 4.3. Survival probability of patients with positive blood cultures and prior
antibiotic use, stratified by antibiotic elimination route

0 500 1000 1500 2000
FU_DAYS
0.0
0.2
0.4
0.6
0.8
1.0
Survival Probability
Through liver Through kidneys exposure
Product-Limit Survival Estimates
With 95% Hall-Wellner Bands
0 500 1000 1500 2000
FU_DAYS
0.0
0.2
0.4
0.6
0.8
1.0
Survival Probability
Through liver Through kidneys exposure
Logrank p <.0001
Product-Limit Survival Estimates
With 95% Hall-Wellner Bands

169

4.7 REFERENCES
1. Goto, M. and M.N. Al-Hasan, Overall burden of bloodstream infection and
nosocomial bloodstream infection in North America and Europe. Clin
Microbiol Infect, 2013. 19(6): p. 501-9.
2. National Nosocomial Infections Surveillance, National Nosocomial
Infections Surveillance (NNIS) System Report, data summary from
January 1992 through June 2003, issued August 2003. Am J Infect
Control, 2003. 31(8): p. 481-98.
3. Heron, M., Deaths: Leading Causes for 2014. Natl Vital Stat Rep, 2016.
65(5): p. 1-96.
4. Pittet, D., et al., Nosocomial bloodstream infection in critically iii patients:
Excess length of stay, extra costs, and attributable mortality. JAMA
271(20): 1598-1601.
5. Al-Rawajfah, O.M., et al., Length of stay and charges associated with
health care-acquired bloodstream infections. Am J Infect Control, 2012.
40(3): p. 227-32.
6. Kaye, K.S., et al., Effect of nosocomial bloodstream infections on
mortality, length of stay, and hospital costs in older adults. J Am Geriatr
Soc, 2014. 62(2): p. 306-11.
7. Wisplinghoff, H., et al., Nosocomial bloodstream infections in US
hospitals: analysis of 24,179 cases from a prospective nationwide
surveillance study. Clin Infect Dis, 2004. 39(3): p. 309-17.

170

8. Bodro, M., et al., Epidemiology, antibiotic therapy and outcomes of
bacteremia caused by drug-resistant ESKAPE pathogens in cancer
patients. Support Care Cancer, 2014. 22(3): p. 603-10.
9. Vazquez-Guillamet, M.C., et al., Predicting Resistance to Piperacillin-
Tazobactam, Cefepime and Meropenem in Septic Patients With
Bloodstream Infection Due to Gram-Negative Bacteria. Clin Infect Dis,
2017. 65(10): p. 1607-1614.
10. Alexis Tabah, e.a., Characteristics and determinants of outcome of
hospital-acquired bloodstream infections in intensive care units: the
EUROBACT International Cohort Study. 2012.
11. Rice, L.B., Federal funding for the study of antimicrobial resistance in
nosocomial pathogens: no ESKAPE. J Infect Dis, 2008. 197(8): p. 1079-
81.
12. Boucher, H.W., et al., Bad bugs, no drugs: no ESKAPE! An update from
the Infectious Diseases Society of America. Clin Infect Dis, 2009. 48(1): p.
1-12.
13. Jakobsson, H.E., et al., Short-term antibiotic treatment has differing long-
term impacts on the human throat and gut microbiome. PLoS One, 2010.
5(3): p. e9836.
14. Dethlefsen, L., et al., The pervasive effects of an antibiotic on the human
gut microbiota, as revealed by deep 16S rRNA sequencing. PLoS Biol,
2008. 6(11): p. e280.

171

15. Dethlefsen, L. and D.A. Relman, Incomplete recovery and individualized
responses of the human distal gut microbiota to repeated antibiotic
perturbation. Proc Natl Acad Sci U S A, 2011. 108 Suppl 1: p. 4554-61.
16. Jernberg, C., et al., Long-term ecological impacts of antibiotic
administration on the human intestinal microbiota. ISME J, 2007. 1(1): p.
56-66.
17. Los Angeles County + University of Southern California Medical Center,
LAC+USC Medical Center Rules and Regulations, 2014.
18. Clinical and Laboratory Standards Institute, Performance Standards for
Antimicrobial Susceptibility Testing, CLSI supplement M100.
19. Magiorakos, A.P., et al., Multidrug-resistant, extensively drug-resistant and
pandrug-resistant bacteria: an international expert proposal for interim
standard definitions for acquired resistance. Clin Microbiol Infect, 2012.
18(3): p. 268-81.
20. Jernberg, C., et al., Long-term impacts of antibiotic exposure on the
human intestinal microbiota. Microbiology, 2010. 156(Pt 11): p. 3216-23.
21. Perez-Cobas, A.E., et al., Differential effects of antibiotic therapy on the
structure and function of human gut microbiota. PLoS One, 2013. 8(11): p.
e80201.
22. Zhang, L., et al., Antibiotic administration routes significantly influence the
levels of antibiotic resistance in gut microbiota. Antimicrob Agents
Chemother, 2013. 57(8): p. 3659-66.

172

23. Chantziaras, I., et al., Studying the effect of administration route and
treatment dose on the selection of enrofloxacin resistance in commensal
Escherichia coli in broilers. J Antimicrob Chemother, 2017. 72(7): p. 1991-
2001.
24. Ghibellini, G., E.M. Leslie, and K.L. Brouwer, Methods to evaluate biliary
excretion of drugs in humans: an updated review. Mol Pharm, 2006. 3(3):
p. 198-211.

173

4.8 SUPPLEMENTARY MATERIAL
Supplementary Table 4.1. Classification of Antibiotics Eliminated through the
Kidneys or through the Liver

Through the Kidneys At least Partially Through the Liver
Cefazolin Metronidazole
Piperacillin and Tazobactam Clindamycin
Vancomycin Erythromycin
Levofloxacin Oxacillin
Gentamicin Ceftriaxone
Cefotaxime Azithromycin
Ceftazidime Linezolid
Ampicillin and Sulbactam Clarithromycin
Imipenem Doxycycline
Ampicillin Chloramphenicol
Sulfamethoxazole and Trimethoprim Trovafloxacin
Ofloxacin Sulfadiazine
Tobramycin Quinupristin and Dalfopristin
Aztreonam Benzathine Penicillin
Cefepime Nitrofurantoin
Amikacin Tetracycline
Penicillin Polymyxin
Ciprofloxacin Atovaquone
Cefuroxime Moxifloxacin
Piperacillin Minocycline
Meropenem Paromomycin
Amoxicillin Alatrofloxacin
Cefotetan Dactinomycin
Cefalexin Mafenide
Cefoxitin Nafcillin
Streptomycin

Ertapenem

Amoxicillin Clavulanate

Imipenem and Cilastatin

Colistimethate

Netilmicin

Ceftizoxime

Cefapirin

Methenamine

Trimethoprim

Cefaclor

Cefixime

Ticarcillin and Clavulanate

174

Supplementary Table 4.1. continued

Through the Kidneys At least Partially Through the Liver
Carbapenem

Norfloxacin

Sulfamethoxazole

Tazobactam

Tobramycin and Vancomycin

175

Supplementary Table 4.2. Association between Antibiotic Elimination Route and
Degree of Resistant Bacteremia in Participating Patients, by Select Stratification
Covariates, 1995-2002

Resistance Score

Route
Organism with no
resistance
(N=451)
Organism with MDR
resistance
†
(N=181)
Organism with XDR
resistance
††
(N=61)
OR (95% CI) OR (95% CI) OR (95% CI)

Gram Status
Positive N=232 N=112 N=12
Through kidneys Ref Ref Ref
Through liver 1.0 (Ref) 1.37 (0.82, 2.28) 6.65 (0.82, 54.19)
p-value -- 0.2266 0.0769

Negative N=219 N=69 N=49
Through kidneys Ref Ref Ref
Through liver 1.0 (Ref) 1.25 (0.69, 2.25) 1.53 (0.75, 3.14)
p-value -- 0.4648 0.2438

Age
≥50 N=188 N=87 N=23
Through kidneys Ref Ref Ref
Through liver 1.0 (Ref) 1.27 (0.73, 2.22) 2.13 (0.71, 6.35)
p-value -- 0.4004 0.1770

<50 N=263 N=94 N=38
Through kidneys Ref Ref Ref
Through liver 1.0 (Ref) 1.48 (0.88, 2.49) 1.50 (0.68, 3.28)
p-value -- 0.1385 0.3146

Gender
Female N=155 N=61 N=21
Through kidneys Ref Ref Ref
Through liver 1.0 (Ref) 1.46 (0.77, 2.77) 1.74 (0.62, 4.85)
p-value -- 0.2503 0.2914

Male N=296 N=120 N=40
Through kidneys Ref Ref Ref
Through liver 1.0 (Ref) 1.32 (0.82, 2.12) 1.81 (0.79, 4.11)
p-value -- 0.2601 0.1583

ICU Category
Surgical ICU N=237 N=87 N=25
Through kidneys Ref Ref Ref
Through liver 1.0 (Ref) 1.26 (0.72, 2.22) 0.81 (0.29, 2.25)
p-value -- 0.4146 0.6890
Odds ratios are reported from multinomial logistic regression adjusted for age and duration in
the ICU using organism with no resistance as the reference group for resistance score.
N=693 patients were included in the analysis.
a
Only Black and Hispanic races are displayed due to small sample size.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.
†††
Any resistance is defined as MDR or XDR resistance.

176

Supplementary Table 4.2. continued
Resistance Score

Route
Organism with no
resistance
(N=451)
Organism with MDR
resistance
†
(N=181)
Organism with XDR
resistance
††
(N=61)
OR (95% CI) OR (95% CI) OR (95% CI)

Medical ICU N=186 N=88 N=33
Through kidneys Ref Ref Ref
Through liver 1.0 (Ref) 1.23 (0.69, 2.19) 1.61 (0.64, 4.01)
p-value -- 0.4874 0.3107

Race
a

Black N=70 N=35 N=17
Through kidneys Ref Ref Ref
Through liver 1.0 (Ref) 1.23 (0.50, 3.03) 0.83 (0.27, 2.57)
p-value -- 0.6502 0.7429

Hispanic N=249 N=88 N=24
Through kidneys Ref Ref Ref
Through liver 1.0 (Ref) 1.08 (0.64, 1.83) 1.69 (0.61, 4.67)
p-value -- 0.7791 0.3162

Odds ratios are reported from multinomial logistic regression adjusted for age and duration in
the ICU using organism with no resistance as the reference group for resistance score.
N=693 patients were included in the analysis.
a
Only Black and Hispanic races are displayed due to small sample size.
†
Multi-drug Resistance (MDR) defined as resistance to ≥3 antimicrobial classes.
††
Extensive-drug Resistance (XDR) defined as resistance to all but ≤2 antimicrobial classes.
†††
Any resistance is defined as MDR or XDR resistance.

177

Supplementary Figure 4.1. Time to blood culture among patients with positive
blood culture and prior antibiotic use, stratified by antibiotic elimination route

0 500 1000 1500 2000
FU_BC
0.0
0.2
0.4
0.6
0.8
1.0
Survival Probability
Through liver Through kidneys exposure
Product-Limit Survival Estimates
With 95% Hall-Wellner Bands
0 500 1000 1500 2000
FU_BC
0.0
0.2
0.4
0.6
0.8
1.0
Survival Probability
Through liver Through kidneys exposure
Logrank p=0.4271
Product-Limit Survival Estimates
With 95% Hall-Wellner Bands

178

Chapter 5: Conclusion

179

5.1 FINDINGS
Antibiotic use has been implicated with subsequent antibiotic resistance in
microbiological, animal, and epidemiological studies [1-9]. In this study, we
reported for the first time a striking pattern of drug resistance associated with
antibiotic use, both in the overall cohort of patients in the ICU (N=11,576), and in
a stringently selected, clinically relevant cohort of patients with antibiotic use and
positive blood cultures (N=693). Our results showed an elevated magnitude of
association between antibiotic use and drug resistance (OR: 2.59 [95% CI: 1.65-
4.05], p<0.0001). We found that drug resistance varied by route of antibiotic
elimination (kidney or liver). Both routes of elimination had elevated effect
estimates for multi-drug and extensive-drug resistance, although patients with
antibiotics eliminated at least partially through the liver had the largest
magnitudes of association. Regardless of which group of patients was selected,
our results consistently showed that elimination of antibiotics through the liver
and biliary tract presented greater odds of resistance compared to elimination of
antibiotics through the kidneys. This trend of increased magnitude of association
for patients with antibiotics eliminated through the liver was also observed in
patients with the ESKAPE organisms.
Sensitivity analyses in which we limited patients to those with acute
antibiotic exposure (within 30 days) preceding the blood culture showed
increased odds of drug resistance among patients with antibiotics eliminated
through the liver, with odds of extensive drug resistance being markedly higher.
This supports the hypothesis that acute antibiotic exposure to the gut can

180

drastically modify the bacterial resistance profile. Patients with any antibiotics
eliminated through the liver administered 31-60 days prior to the blood culture
also had elevated odds of resistance, although the effect estimates were
attenuated. We found that patients with antibiotics eliminated through the liver
had poorer survival in the hospital (p<0.0001) compared to patients with
antibiotics eliminated through the kidneys. Finally, 61.3% of drug resistance in
our cohort was attributable to antibiotic use.
5.2 STRENGTHS AND LIMITATIONS
A strength of our studies is the use of a large, ethnically-diverse hospital-
based cohort that spans 1995 to 2002. Another strength of our studies is the
ability to control for confounding of numerous patient and hospital characteristics,
including sex, gender, race, type of admitting diagnosis, comorbidities, and
organism gram stain classification. Additionally, we have a consistent set of
findings which may be generalized to patients in the ICUs of US hospitals that
have patient characteristics similar to those of patients in our cohort.
One limitation of our studies was the broad measurement of the exposure
variable. For each antibiotic, the exposure measure was classified dichotomously
as either predominantly through the kidneys or at least partially through the liver
based on whether dose adjustment for the antibiotic was needed for patients with
compromised renal function. However, it is likely that among antibiotics whose
parent drug and metabolites eliminated primarily through the kidneys, a small
proportion are eliminated through the gut. Therefore, it is possible that even low
antibiotic exposure to the gut of antibiotics eliminated predominantly through the

181

kidneys is enough to impact drug resistance. This may explain why we have
observed patients with antibiotics eliminated through the kidneys with elevated
risk of drug resistance. Any misclassification of the exposure is likely to be non-
differential, and expected to bias estimates of the association of route of
antibiotic elimination and drug resistance towards the null value. Our definition of
exposure would not explain the strong association between elimination route and
drug resistance found in our studies.
An additional limitation of our study is the potential for confounding by
indication, in which the apparent association between the association between
route of elimination and drug resistance could be due to an underlying but
unaccounted-for variable. Assessment of admitting diagnoses and selected
comorbidities in the baseline period demonstrated differing distributions by the
route of antibiotic elimination, indicating that patients could have had differing
severity of disease for antibiotics eliminated through the kidneys versus through
the liver. However, we addressed the possibility of confounding by indication by
controlling for these admitting diagnoses and selected comorbidities, along with
age, type of ICU, ICU LOS, sex, and gram stain classification, in our analyses.
Within the clinically relevant cohort, we also used the time from admission into
the ICU to the blood draw of the qualifying organism for patients with antibiotics
eliminated through the kidneys versus patients with through the liver as an
indicator for differing severity of disease. We found that the proportion of patients
diagnosed with an HA-BSI at any given time during their ICU stay did not differ
by elimination route. Nevertheless, it is possible that there is some other

182

underlying indication influencing both the route of antibiotic elimination and drug
resistance.
5.3 FUTURE DIRECTIONS
Further research to address the limitations of our studies is needed. To
measure the route of antibiotic elimination in a quantitative manner, we may
assess the association between deciles of antibiotic and metabolite elimination in
the gut and drug resistance. However; additional studies need to be conducted to
more precisely measure the distribution of antibiotics and their metabolites in the
urine and feces. A randomized clinical trial can address the possibility of
confounding by indication beyond the methods implemented in our studies.
Patients would be randomized to standards of care for antibiotics eliminated
predominately through the kidneys or antibiotics eliminated at least partially
through the liver, based on site of infection and other patient characteristics, and
subsequently followed-up for a HA-BSI organism. If further research
substantiates findings from our studies, then limiting exposure of antibiotics in the
gut may be a novel strategy for antibiotic resistance control within the hospital
setting. Furthermore, this may warrant the deliberate designing of the next
generation of antibiotics to be eliminated only through the kidneys.

183

5.4 REFERENCES
1. Jernberg, C., et al., Long-term impacts of antibiotic exposure on the
human intestinal microbiota. Microbiology, 2010. 156(Pt 11): p. 3216-23.
2. Jakobsson, H.E., et al., Short-term antibiotic treatment has differing long-
term impacts on the human throat and gut microbiome. PLoS One, 2010.
5(3): p. e9836.
3. Perez-Cobas, A.E., et al., Differential effects of antibiotic therapy on the
structure and function of human gut microbiota. PLoS One, 2013. 8(11): p.
e80201.
4. Jernberg, C., et al., Long-term ecological impacts of antibiotic
administration on the human intestinal microbiota. ISME J, 2007. 1(1): p.
56-66.
5. Dethlefsen, L., et al., The pervasive effects of an antibiotic on the human
gut microbiota, as revealed by deep 16S rRNA sequencing. PLoS Biol,
2008. 6(11): p. e280.
6. Dethlefsen, L. and D.A. Relman, Incomplete recovery and individualized
responses of the human distal gut microbiota to repeated antibiotic
perturbation. Proc Natl Acad Sci U S A, 2011. 108 Suppl 1: p. 4554-61.
7. Nyberg, S.D., et al., Long-term antimicrobial resistance in Escherichia coli
from human intestinal microbiota after administration of clindamycin.
Scand J Infect Dis, 2007. 39(6-7): p. 514-20.

184

8. Zhang, L., et al., Antibiotic administration routes significantly influence the
levels of antibiotic resistance in gut microbiota. Antimicrob Agents
Chemother, 2013. 57(8): p. 3659-66.
9. Chantziaras, I., et al., Studying the effect of administration route and
treatment dose on the selection of enrofloxacin resistance in commensal
Escherichia coli in broilers. J Antimicrob Chemother, 2017. 72(7): p. 1991-
2001.

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Investigating a physiological pathway for the effect of guided imagery on insulin resistance

PDF

Vision epidemiology and the impact of vision loss on vision-specific quality of life

PDF

Body size and the risk of prostate cancer in the multiethnic cohort

PDF

Age related macular degeneration in Latinos: risk factors and impact on quality of life

PDF

Chronic eye disease epidemiology in the multiethnic ophthalmology cohorts of California study

PDF

Validation of the pancreatitis activity scoring system in a large prospective cohort

PDF

Methicillin-resistant Staphylococcus aureus: molecular epidemiology, virulence in disease, and antibiotic modulation of virulence

PDF

The environmental and genetic determinants of cleft lip and palate in the global setting

PDF

Lymph node metastases in non-rhabdomyosarcoma soft tissue sarcoma

Asset Metadata

Creator Aggarwal, Shivani (author)

Core Title Antibiotic resistance in a large ICU cohort: 1995-2002

School Keck School of Medicine

Degree Doctor of Philosophy

Degree Program Epidemiology

Publication Date 05/13/2019

Defense Date 12/05/2018

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag antibiotic elimination route,antibiotic resistance,bloodstream infection,ESKAPE,extensive-drug resistance,gut microbiome,hospital-acquired bloodstream infection,multi-drug resistance,OAI-PMH Harvest

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Cortessis, Victoria (committee chair), Cozen, Wendy (committee member), Mack, Thomas (committee member), McKean-Cowdin, Roberta (committee member), Minejima, Emi (committee member)

Creator Email shivani.s.aggarwal@gmail.com,shivania@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-170291

Unique identifier UC11660551

Identifier etd-AggarwalSh-7390.pdf (filename),usctheses-c89-170291 (legacy record id)

Legacy Identifier etd-AggarwalSh-7390.pdf

Dmrecord 170291

Document Type Dissertation

Format application/pdf (imt)

Rights Aggarwal, Shivani

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA