E D I T O R I A L C O M M E N TA R Y

Looking for Risk Factors for the Acquisition

of Antibiotic Resistance: A 21st-Century Approach
David L. Paterson
Antibiotic Management Program, Division of Infectious Diseases, University of Pittsburgh Medical Center, Pennsylvania

(See the article by Harris et al. on pages 155863.)

Received 19 February 2002; electronically published 23

May 2002.
Reprints or correspondence: Dr. David L. Paterson,
Antibiotic Management Program, Ste. 3A, Falk Medical Bldg.,
3601 5th Ave., Pittsburgh, PA 15213 (patersond@msx.deptmed.pitt.edu).
Clinical Infectious Diseases 2002; 34:15647
 2002 by the Infectious Diseases Society of America. All
rights reserved.
1058-4838/2002/3412-0004$03.00

tors for the selection of antibiotic resistance can be variously made. Examples
include (1) patients with any clinical specimen exhibiting the resistance phenotype
under study, (2) patients with unequivocally significant isolates (e.g., patients with
bacteremia), and (3) patients with an infection defined by clinical criteria (thereby
excluding patients whose isolates represent
colonization rather than true infection).
Each of these definitions has limitations. For each patient from whom a clinical specimen is obtained that harbors an
antibiotic-resistant organism, there may
be 25 other patients with asymptomatic
colonization (especially of the gastrointestinal tract). Colonized patients may be erroneously classified as control patients,
thereby leading to a potential underestimation of the risk provided by use of certain antibiotics. Therefore, it could be argued that, in an ideal case-control study
that examines the risk factors for selection
of antibiotic resistance, case patients
would be defined as all patients colonized
with the antibiotic-resistant organism under study (table 1). In practice, such an
approach would require systematic prospective surveillance for resistant organisms (e.g., by periodic rectal or nose swabs
of at-risk patients).
The acquisition of antibiotic-resistant
bacteria within our hospitals represents
the combined result of the selection of re-

1564 CID 2002:34 (15 June) EDITORIAL COMMENTARY

sistant organisms via antibiotic use and

horizontal transfer, usually via the hands
of health care workers, but sometimes via
inanimate objects. In virtually all casecontrol studies, patients with both modes
of acquisition of the antibiotic-resistant
organism are lumped together, regardless
of the mode of acquisition of the organism. This may be erroneous, because antibiotic pressure may play a lesser role
in those patients who have acquired the
organism by horizontal transfer. Modern molecular epidemiologic tools (e.g.,
pulsed-field gel electrophoresis or PCR
based on repetitive chromosomal sequences) can allow patients who have acquired organisms from other patients to
be distinguished from patients whose antibiotic therapy has resulted in the selection of resistant organisms.
However, recent data from Donskey et
al. [1] have shown that the 2 groups may
not be mutually exclusive (figure 1). Antibiotics with antianaerobic activity increased the fecal density of vancomcyinresistant enterococci (VRE) in patients
already known to be colonized with VRE.
Those patients with a high fecal density of
VRE were more likely to have environmental contamination with the organism.
Therefore, it could be supposed that these
patients would be more likely to serve as
sources for horizontal transfer of VRE
than would patients with lower fecal den-

Downloaded from http://cid.oxfordjournals.org/ by guest on August 18, 2014

A popular method of examining the risk

factors for the acquisition of antibioticresistant organisms is the case-control
study. More than 100 such studies addressing antimicrobial resistance have now
been published. These studies have yielded
a plethora of conflicting resultsin some
cases, this may reflect differences in local
epidemiology, but, in other cases, methodological differences can be identified as
the root cause of the diversity of results.
The time has now come to approach studies that assess risk factors for the selection
of antibiotic resistance with a consistency
of methods and a renewed sense of the
true purpose of such studies. They should
not simply be projects that can be accomplished by a clinically oriented infectious
diseases fellow during the duration of his
or her fellowship, but, rather, studies
that aim to change antibiotic-prescribing
patterns and to demonstrate that such
changes improve outcomes for patients.
The definition of case patients for a
case-control study that examines risk fac-

Table 1.

The ideal case-control study for examination of the risk factors for the acquisition of antibiotic-resistant organisms.

Variable

Ideal study

Usual study

Selection of case patients

All patients colonized or infected with the antibioticresistant organism under study

All patients with clinical isolates of the antibiotic-resistant organism under study

Selection of control patients

Patients drawn from the same population as case patients

All patients with susceptible forms of the

organism under study

Same duration of exposure as case patients

Varying durations of exposure

Similar severity of illness as case patients

Varying severities of illness

Determination of resistance mechanisms

Separate analysis of patients with genotypically distinct

organisms vs. those with genotypically similar
organisms

Single analysis of all patients lumped together

Molecular epidemiological
analysis

Separate analysis of patients with organisms with different resistance mechanisms

Single analysis of all patients lumped together

Intervention aimed at modifying use of the implicated

antibiotic, with assessment of effect of intervention
on outcomes for patients (mortality, length of hospital
stay, and costs of antimicrobial use)

No intervention

Analyses

Intervention

resistant organisms. Most case-control

studies have concentrated on phenotypic resistance. For example, Harris et
al. [2] have studied imipenem resistance
in Pseudomonas aeruginosa. However, the
phenotype for imipenem resistance may
be the result of a variety of different mechanismsfor example, loss of the outermembrane protein OprD, production of
carbapenemases of various types, or upregulation of the efflux system MexEFOprN [3]. Similarly, the phenotype for
ceftazidime resistance in Klebsiella pneumoniae may be the result of production
of a group 2be b-lactamase (an extendedspectrum b-lactamase) or a group 1 blactamase (AmpC-type enzyme) or of
hyperproduction of a group 2b b-lactamase in the presence of loss of an outermembrane protein [4].
It is likely that different antibiotics predispose to different mechanisms of antibiotic resistance. Whether this will be of
great clinical importance has yet to be determined. However, sophisticated casecontrol studies could assess differences in
antibiotic risk factors for organisms with
the same resistance phenotype but different mechanisms of antibiotic resistance.
Harris, Samore, Kaye, and Carmeli have
been leaders in the refinement of selection
of control groups in risk-factor studies of

antibiotic-resistant organisms. They have

now produced a series of papers [2, 57]
that have emphasized a central theme: that
case-control studies that compare patients
who have the resistant form of an organism (case patients) with those who have
the susceptible form of an organism (control patients) may overestimate the contribution of the resistance-defining antibiotic in the development of resistance.
Additionally, as they illustrate in the article
on which I comment, some antibiotics
may be falsely implicated as potential risk
factors for acquisition by use of the traditional resistant p case, susceptible p
control design.
An additional potential cause of bias in
some studies with the resistant p case,
susceptible p control design is the inclusion of patients who have communityacquired infections along with patients
who have nosocomial infections. For example, a study of extended-spectrum
cephalosporin resistance in Enterobacteriaceae that includes both communityacquired and nosocomial infections is
highly likely to show an association between extended-spectrum cephalosporin
resistance and the use of parenterally administered antibiotics, such as vancomycin. Use of vancomycin is unlikely to
be a biologically plausible risk factor for

EDITORIAL COMMENTARY CID 2002:34 (15 June) 1565

Downloaded from http://cid.oxfordjournals.org/ by guest on August 18, 2014

sities. To my knowledge, it is not known

whether certain antibiotics increase the
propensity of horizontal transfer for other
resistant bacteria. Case-control studies
performed solely on those patients colonized with a single clone could potentially
raise hypotheses regarding the effects of
antibiotics on the propensity for horizontal transmission of resistant organisms.
Optimally, a separate analysis could be
performed on case patients with genotypically diverse strainssuch an analysis
would assess the risk factors for de novo
selection of resistant organisms. Existing
data have suggested that some antibiotics
may produce effects in patients already
colonized with antibiotic-resistant organisms that differ from the effects in those
who have never been colonized. Thus, use
of a particular antibiotic may be a risk
factor for increasing fecal density of resistant organisms in patients who are already
colonized (and, therefore, may facilitate
patient-to-patient spread) and yet may
even prevent de novo selection of the same
resistant organisms in noncolonized patients. This concept needs to be further
explored by appropriately designed casecontrol studies.
Other modern molecular biological
tools can further enhance our study of the
epidemiologic characteristics of antibiotic-

Figure 1. Acquisition of antibiotic-resistant bacteria from either in vivo selection by antibiotic use or horizontal transmission of genotypically
identical organisms. An unknown component of horizontal transmission is facilitated by antibiotics that increase the density of colonization with
resistant organisms.

studies that have assessed antibioticresistant organisms are the Acute Physiology and Chronic Health Evaluation
(APACHE) III score [8] and the Pitt bacteremia index [9]. It should be noted that
the APACHE III score is only applicable to patients in ICUs, and, although
it is a predictor of mortality, it may not
necessarily be a predictor of colonization or infection with antibiotic-resistant
organisms.
Given these comments about the selection of case and control patients (summarized in table 1), how should we apply
the results of case-control studies that explore risk factors for acquisition of antibiotic-resistant organisms? Kollef et al.
[10] have recently demonstrated that the
use of inadequate empiric antibiotic therapy in patients hospitalized in ICUs was
associated with increased mortality. Inadequate empiric antibiotic therapy was
most frequently caused by unsuspected
antibiotic resistance. Therefore, intuitively,
it seems right to do all we can to prevent
antibiotic resistance, even if it means restriction of the use of certain antibiotics.
However, we should put such policies in
practice only after evaluation of the big
picture. The adverse effects on the outcome of antibiotic resistance in one organism need to be compared with the adverse effects of increased resistance in
other organisms if antibiotic-use patterns

1566 CID 2002:34 (15 June) EDITORIAL COMMENTARY

change. Will restriction of the use of one

antibiotic lead to an increase in use of another antibiotic that may have greater deleterious effects?
Optimally, the results of case-control
studies that assess risk factors for the selection of antibiotic resistance should be
applied to interventions aimed at reducing
the selection of antibiotic resistance in the
organism under study. The results of such
interventions should be published with
the results of the case-control study. This
was the problem. This is what we did
about it. This was the result of our intervention. Such results should illustrate effects on not just the antibiotic-resistance
issue under study but also on the outcomes for patients (i.e., mortality, length
of hospital stay, and costs of antimicrobial
use). Many hospitals are increasing control over antibiotic prescribing by creating
antibiotic-management programs [11].
These provide ideal vehicles with which to
put modifications of antibiotic use into
practice, with an aim being to optimize
the choice of antimicrobial agents. The results of such programs will be the ultimate
arbiter of the usefulness of the search for
the risk factors for antibiotic resistance.

References
1. Donskey CJ, Chowdhry TK, Hecker MT, et al.
Effect of antibiotic therapy on the density of

Downloaded from http://cid.oxfordjournals.org/ by guest on August 18, 2014

the selection of cephalosporin resistance

in Enterobacteriaceae. Extended-spectrum cephalosporinresistant Enterobacteriaceae are more likely to be found in
nosocomial infections, and parenterally
administered antibiotics are more likely to
be used in hospitals, which confounds the
original observation.
In a study of the risk factors for nosocomial infection with antibiotic-resistant bacteria, the optimal control group
would be hospitalized patients with the
same potential exposure to the antibioticresistant bacteria as case patients. Thus,
control patients would not only be hospitalized during the same period as case
patients, but they would also be hospitalized for at least the same duration as the
time from admission of the case patient
to the first isolation of the resistant organism in question. A general principle is
that control patients should be from the
same source population as case patients;
patients in rehabilitation units attached to
tertiary care hospitals are not likely to be
appropriate control subjects for case patients who are primarily hospitalized in
intensive care units (ICUs), regardless of
whether they meet the definition for the
same duration of stay. Severity of illness
is yet another potential confounding variablenumerous scoring systems have
been developed that can quantify severity
of illness. Among the most widely used in

vancomycin-resistant enterococci in the stool

of colonized patients. N Engl J Med 2000; 343:
192532.
2. Harris AD, Smith D, Johnson JA, Bradham
DD, Roghmann MC. Risk factors for imipenem-resistant Pseudomonas aeruginosa among
hospitalized patients. Clin Infect Dis 2002; 34:
3405.
3. Livermore DM. Multiple mechanisms of antimicrobial resistance in Pseudomonas aeruginosa: our worst nightmare? Clin Infect Dis
2002; 34:63440.
4. Rice LB, Carias LL, Hujer AM, et al. Highlevel expression of chromosomally encoded
SHV-1 b-lactamase and an outer membrane
protein change confer resistance to ceftazidime and piperacillin/tazobactam in a clinical
isolate of Klebsiella pneumoniae. Antimicrob

Agents Chemother 2000; 44:3627.

5. Harris AD, Samore MH, Lipsitch M, Kaye KS,
Perencevich E, Carmeli Y. Control-group selection importance in studies of antimicrobial
resistance: examples applied to Pseudomonas
aeruginosa, enterococci, and Escherichia coli.
Clin Infect Dis 2002; 34:155863 (in this
issue).
6. Harris AD, Karchmer TB, Carmeli Y, Samore
MH. Methodological principles of casecontrol studies that analyzed risk factors for
antibiotic resistance: a systematic review. Clin
Infect Dis 2001; 32:105561.
7. Kaye KS, Harris AD, Gold H, Carmeli Y. Risk
factors for recovery of ampicillin-sulbactamresistant Escherichia coli in hospitalized
patients. Antimicrob Agents Chemother
2000; 44:10049.

8. Knaus WA, Wagner DP, Draper EA, et al. The

APACHE III prognostic system: risk prediction of hospital mortality for critically ill hospitalized adults. Chest 1991; 100:161936.
9. Chow JW, Yu VL. Combination antibiotic
therapy versus monotherapy for gram-negative bacteremia: a commentary. Int J Antimicrob Agents 1999; 11:712.
10. Kollef MH, Sherman G, Ward S, Fraser VJ.
Inadequate antimicrobial treatment of infections: a risk factor for hospital mortality
among critically ill patients. Chest 1999; 115:
46274.
11. Fishman NO. Antimicrobial management and
cost containment. In: Mandell GL, Bennett JE,
Dolin R, eds. Principles and practice of infectious diseases. 5th ed. Philadelphia: Churchill
Livingstone, 2000:53946.

Downloaded from http://cid.oxfordjournals.org/ by guest on August 18, 2014

EDITORIAL COMMENTARY CID 2002:34 (15 June) 1567