American Journal of Infection Control 44 (2016) 134-7

Contents lists available at ScienceDirect

American Journal of Infection Control American Journal of

Infection Control

journal homepage: www.ajicjournal.org

Major article

Demographic and infection characteristics of patients with

carbapenem-resistant Enterobacteriaceae in a community hospital:
Development of a bedside clinical score for risk assessment
Brooke M. Miller PharmD a, Steven W. Johnson PharmD, BCPS a, b, *
a
Novant Health Forsyth Medical Center, Winston-Salem, NC
b
Campbell University College of Pharmacy and Health Sciences, Buies Creek, NC

Key Words: Background: The objective of this study was to identify risk factors associated with the presence of
Carbapenem-resistant Enterobacteriaceae carbapenem-resistant Enterobacteriaceae (CRE) infections to develop a clinical prediction model that can
CRE be used at patient bedside to identify subjects likely infected with a CRE pathogen.
Prediction model
Methods: This case-control study included patients aged
18 years admitted to Novant Health Forsyth
Risk factors
Medical Center between January 1, 2012, and December 31, 2013, with CRE infections (cases) or non-CRE
Enterobacteriaceae
infections (controls). Controls were matched to their corresponding resistant case (3:1) based on path-
ogen, place of likely acquisition, isolate source, year of admission, and level of care. A risk prediction
model was developed using variables independently associated with CRE isolation. Sensitivities and
specificities were obtained at various point cutoffs, and a determination of the receiver operator char-
acteristic (ROC) area under the curve (AUC) was performed.
Results: A total of 164 subjects were included. Independent risk factors for CRE included recent antibiotic
therapy, recent immunosuppression, and Charlson Comorbidity Index score
4. Adjusted odds ratios
were 13.37 (95% confidence interval [CI], 4.16-61.19), 6.69 (95% CI, 1.85-29.65), and 3.30 (95% CI,
1.34-8.40), respectively. Diagnostic performance of various score cutoffs for the model indicated a score

5 correlated with the highest accuracy (79%). The ROC AUC was 0.83.
Conclusion: The risk prediction model displayed good discrimination and was an excellent predictor of
CRE infection.
Copyright Ó 2016 by the Association for Professionals in Infection Control and Epidemiology, Inc.
Published by Elsevier Inc. All rights reserved.

Carbapenem-resistant Enterobacteriaceae (CRE) have been
categorized as an urgent threat level by the Centers for Disease
Control and Prevention (CDC). The CDC estimated that there were
approximately 9,300 CRE infections resulting in approximately 600
deaths in 2013.1 The most common types of CRE include
carbapenem-resistant Klebsiella spp, accounting for 11% of Enter-
obacteriaceae health careeassociated infections, and carbapenem-
resistant Escherichia coli, accounting for 2% of Enterobacteriaceae
health careeassociated infections.1 Patients infected with CRE are
* Address correspondence to Steven W. Johnson, PharmD, BCPS, Novant Health
Forsyth Medical Center, 3333 Silas Creek Pkwy, Winston-Salem, NC 27103.
E-mail address: johnsonsw@campbell.edu (S.W. Johnson).
Previous presentations: This research was previously presented in full as a
podium presentation at the Southeastern Residency Conference, April 30, 2015,
Athens, GA; and in part as an abstract poster presentation at the American Society
of Health-Systems Pharmacy Midyear Clinical Meeting, December 9, 2014, Ana-
heim, CA.
Conflicts of interest: None to report.
subject to delayed appropriate initial antibiotic therapy (IAT) and,
consequently, increased mortality.2
To begin appropriate IAT in a timely manner, many health care
institutions have recognized the need for risk stratification tools to
identify patients at increased risk for infection as a result of CRE
organisms. Recent reports have evaluated risk factors for CRE and
have proposed various risk factor scoring models; however, these
models were created with specific patient populations and organ-
isms.2-4 Consequently, generalizability and utility in other settings
is unknown.
One study examined predictors of carbapenem-resistant K
pneumoniae acquisition among hospitalized adults. They found that
poor functional status, intensive care unit stay, and receipt of anti-
biotics, particularly fluoroquinolones, were independent risk fac-
tors for carbapenem-resistant K pneumoniae isolation.3 This study
only identified risk factors for carbapenem-resistant K pneumoniae,
failing to include other potential CRE-producing organisms. A case-
case-control study found that intensive care unit stay and

0196-6553/$36.00 - Copyright Ó 2016 by the Association for Professionals in Infection Control and Epidemiology, Inc. Published by Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.ajic.2015.09.006
 B.M. Miller, S.W. Johnson / American Journal of Infection Control 44 (2016) 134-7 135

cumulative number of antibiotic days were independent risk factors Table 1

of CRE isolation.4 This study included only nosocomial infections, Demographic and infection characteristics of patients with CRE in 2012 and 2013

excluding potential community-acquired CRE. Another study Characteristic 2012 2013 Total
developed a bedside scoring system to differentiate bloodstream Patient characteristics
infections caused by CRE versus extended-spectrum b-lactamasee Age, y 72.3  14.4 63.5  11.2 67.3  13.3
producing Enterobacteriaceae. They identified neurologic disease, Male 5 (28) 14 (61) 20 (46)
CRE pathogen
dependent functional status at admission, diabetes mellitus,
Klebsiella pneumoniae 8 (44) 10 (43) 18 (44)
intensive care unit admission, and antibiotic exposure in 3 months Enterobacter spp 9 (50) 11 (48) 20 (49)
before admission to be risk factors predictive of CRE infection.2 This Citrobacter freundii 1 (6) 0 (0) 1 (2)
study only included patients with severe sepsis, septic shock, or Morganella morganii 0 (0) 1 (4) 1 (2)
multiorgan failure with bloodstream infections, restricting the use Proteus mirabilis 0 (0) 1 (4) 1 (2)
Isolate source
of the scoring system to a small subset of patients. Urinary tract 13 (72) 12 (52) 25 (61)
To our knowledge, no risk prediction model has been published Skin and soft tissue 3 (17) 3 (13) 6 (15)
which evaluates all CRE organisms and all sources of infections in Blood 0 (0) 4 (17) 4 (10)
hospitalized patients. The primary objective of this study was to Lower respiratory tract 2 (11) 4 (17) 6 (15)
CRE susceptibility
identify risk factors associated with the presence of CRE infections
Aminoglycoside 13 (72) 13 (48) 26 (58)
at Novant Health Forsyth Medical Center (NHFMC). The secondary Ciprofloxacin 6 (33) 12 (44) 18 (40)
objective was to develop a clinical risk prediction model that can be Sulfamethoxazole-trimethoprim 7 (39) 15 (56) 22 (49)
used at patient bedside to identify subjects likely infected with a Tigecycline* 1 (50) 3 (43) 4 (44)
CRE pathogen to facilitate IAT. Comorbidities
Charlson Comorbidity Index score 3.6  2.8 3.3  2.0 3.4  2.3
Diabetes 12 (67) 10 (43) 22 (54)
METHODS Chronic respiratory disease 6 (33) 10 (43) 16 (39)
Chronic heart failure 8 (50) 3 (13) 11 (27)
This was a retrospective, single-center, case-control study. Po- Chronic kidney disease 7 (39) 6 (26) 13 (32)
Dementia 5 (28) 2 (9) 7 (17)
tential subjects were identified using the NHFMC infection control
Liver disease 2 (11) 6 (23) 8 (20)
database. Subjects were included if they were aged
18 years and Connective tissue disease 2 (11) 4 (17) 6 (15)
admitted to NHFMC between January 1, 2012, and December 31, Medical history
2013. This study was approved by NHFMC’s Institutional Review Transfer from another health care facility 9 (50) 12 (52) 21 (51)
Board. Cases were defined as patients infected with a CRE obtained Recenty urinary catheterization 10 (56) 15 (65) 25 (61)
Recentz antibiotic therapy 16 (89) 22 (96) 38 (93)
from a clinical culture during admission. If >1 isolate was reported b-lactam therapy 13 (72) 21 (91) 34 (83)
for the same patient, only the index culture was included in the Fluoroquinolone therapy 5 (28) 9 (39) 14 (34)
study. CRE phenotype was based on Clinical Laboratory Standard Carbapenem therapy 4 (22) 5 (22) 9 (22)
Instituteeapproved methods and interpretive criteria.5 Controls Recentx hospitalization 17 (94) 20 (87) 37 (90)
z
Recent immunosuppression 5 (28) 6 (26) 11 (27)
were defined as patients with Enterobacteriaceae (non-CRE) in-
Amputation-paraplegia 3 (17) 3 (13) 6 (15)
fections. Controls were matched to their corresponding resistant Patient admission location
case in a 3:1 fashion based on pathogen, place of likely acquisition Medicine unit 8 (44) 9 (39) 17 (41)
(ie, health care associated vs community acquired), isolate source, Critical care unit 9 (50) 14 (61) 23 (56)
year of admission, and level of care (ie, critical care vs medicine). Other units 1 (6) 0 (0) 1 (2)
Location of likely acquisition of CRE
Subjects were excluded if the positive culture was thought to Hospital or health care associated 17 (94) 20 (87) 37 (90)
represent contamination or colonization (eg, no signs or symptoms Community 1 (6) 3 (13) 4 (10)
of active infection), medical records were incomplete, or a control
NOTE. Values are mean  SD or n (%).
patient had a prior history of CRE. CRE, carbapenem-resistant Enterobacteriaceae.
Data collection from both electronic and paper-based medical *Not all cultures tested for susceptibility.
y
records included patient demographics (age, sex, admission date, Within the last 30 days before index hospitalization.
z
admission unit); culture results; antibiotic therapy during the 3 Within the last 90 days before obtaining index culture.
x
Within the last year before index hospitalization.
months preceding index culture; hospitalization during the
12 months preceding index hospitalization; transfer from another
health care facility; urinary catheterization during the 30 days pre- approach was used to identify risk factors for CRE isolation. Vari-
ceding the index hospitalization; comorbidities included in the ables were kept in the final model if the P value was <.05. The final
Charlson Comorbidity Index6; and immunosuppressive drug therapy regression model was transformed into a point-based tool with
during the 3 months preceding index culture, defined as glucocorti- weighted scores assigned to the variables identified to be associ-
coids (equivalent to prednisone
20 mg for
2 weeks), tacrolimus, ated with CRE isolation. The scores assigned to each variable were
sirolimus, cyclosporine, mycophenolate, or antithymocyte globulin. obtained by dividing each regression coefficient by half of the
Standard descriptive statistics were used to describe the study smallest coefficient and rounding to the nearest integer. The
cohort and compare cases with controls. Continuous variables were sensitivity, specificity, negative predictive value, and positive pre-
expressed as means and were compared using Student t test for dictive value of the prediction tool were expressed at various point
normally distributed variables. Categorical variables were analyzed cutoffs. A receiver operator characteristic area under the curve
by c2 or 2-tailed Fisher exact test. Odds ratio and 95% confidence (AUC) was performed on the model to determine the model’s
intervals were calculated to evaluate the strength of any association predictive value.
that emerged. JMP 8 statistical software (SAS Institute, Cary, NC)
was used to perform the statistical analyses. A univariate analysis RESULTS
was performed to identify variables significantly more associated
with cases than controls (defined as P < .05). Variables associated A total of 164 subjects (41 cases, 123 controls) were included.
(P  .10) with CRE isolation in the univariate analysis were then Table 1 compares CRE cases isolated in 2012 with CRE cases isolated
included into a logistic regression model, and a backward stepwise in 2013. The number of CRE infections increased from 2012 to 2013
136 B.M. Miller, S.W. Johnson / American Journal of Infection Control 44 (2016) 134-7

Table 2 DISCUSSION
Demographic and infection characteristics of patients infected with CRE versus non-
CRE infected
The CDC has classified CRE as an urgent threat because of its
Characteristic Cases Controls P value increasing prevalence, high rates of mortality in infections, and
Patient characteristics difficulty to treat.1 Institutions have recognized the public health
Age, y 67.3  13.3 68.3  15.6 .72 concern associated with various multidrug-resistant organisms,
Male 20 (46) 42 (34) .10
including CRE, and subsequently developed risk prediction
CRE pathogen
Klebsiella pneumoniae 18 (44) 54 (44) tools.2,7-13 To combat the threat of CRE at our institution, we have
Enterobacter spp 20 (49) 60 (49) attempted to develop a risk prediction model to facilitate IAT.
Citrobacter freundii 1 (2) 3 (2) Elements of the CRE scoring model developed in this study are
Morganella morganii 1 (2) 3 (2) similar to risk factors previously identified.2-4,10 In studies eval-
Proteus mirabilis 1 (2) 3 (2)
Isolate source
uating risk factors for CRE infections, recent antibiotic therapy
Urinary tract 25 (61) 75 (61) appears to be the most prominent risk identified. In this study,
Skin and soft tissue 6 (15) 18 (15) recent antibiotic therapy was associated with the highest odds
Blood 4 (10) 12 (10) ratio for CRE isolation. However, other previously identified risk
Lower respiratory tract 6 (15) 18 (15)
factors (eg, transfer from another health care facility) were not
Comorbidities
Charlson Comorbidity Index score 3.4  2.3 2.24  1.95 .009 found to be significant risk factors in the present study. This is
Medical history possibly related to geographic variability in the prominence of
Transfer from another health care facility 21 (51) 44 (36) .098 CRE.
Recent* urinary catheterization 25 (61) 67 (54) .59 The utility of this scoring model and how it should be used are
Recenty antibiotic therapy 38 (93) 54 (44) <.0001
z
Recent hospitalization 37 (90) 80 (65) .001
yet to be determined. The first practical step to applying this
Recenty immunosuppression 11 (27) 4 (3) <.0001 model would be setting a cutoff score. However, the optimal
Patient admission location cutoff score depends on the purpose of the scoring model.7 The
Medicine unit 17 (41) 51 (41) significance of a scoring model is based on the expected utility
Critical care unit 23 (56) 69 (56)
linked with its AUC.14,15 If the scoring model will be used as a
Other units 1 (2) 3 (2)
Location of likely acquisition of CRE screening tool, choosing a cutoff score with a high sensitivity
Hospital or health care associated 37 (90) 111 (90) with the sacrifice of a lower specificity (eg,
2) may be appro-
Community 4 (10) 12 (10) priate. Using the model in this manner would allow withholding
NOTE. Date are n (%), mean  SD, or as otherwise indicated. empirical antibacterial therapy with in vitro activity against CRE
CRE, carbapenem-resistant Enterobacteriaceae. in patients with scores below the cutoff. However, using the
*Within the last 30 days before index hospitalization. model in this manner will likely result in overuse of antibiotics,
y
Within the last 90 days before obtaining index culture.
z especially less desired antibiotics (eg, colistin). This approach
Within the last year before index hospitalization.
may be justified in more severe infections (eg, bacteremia). In
contrast, empirical use of antibiotic therapy directed at CRE
from 18 to 23. Clinical, laboratory, and demographic characteristics would likely require a cutoff score with a higher specificity and
of the study population are summarized in Table 2. Most patients positive predictive value with the sacrifice of a lower sensitivity
were women (54% of cases, 66% of controls), and the median age (eg,
5). Therefore, if only subjects with a score
5 were given
was 67.3 years for cases and 68.3 years for controls. K pneumoniae empirical therapy directed at CRE, 79% of patients would have
and Enterobacter spp were the most common causative microor- received appropriate IAT, 9% would have been treated too
ganisms isolated with 44% and 49%, respectively. The most com- broadly, and 12% would not have been receiving necessary initial
mon source of infection was urine (61%), followed by respiratory empirical CRE-directed therapy. Other potential utility of this
and wound (14.6% and 12.2%, respectively). Most patients were model could be related to infection control and antimicrobial
located in a critical care bed, accounting for 56.1% of subjects, and stewardship practices (eg, guide initiation of empirical contact
most infections were hospital acquired or health care associated, precautions until CRE was ruled out).
accounting for 90% of subjects. Our study had some limitations. Because of the retrospective
The mean Charlson Comorbidity Index score (P ¼ .009), recent design of this study, data were unavailable if not present in the
antibiotic therapy (P < .0001), recent hospitalization (P ¼ .001), and NHFMC medical record. In previous studies, intensive care unit
recent immunosuppressive therapy (P < .0001) were found to be stay was shown to be a risk factor for CRE infection.2-4 This study
significant risk factors for the development of CRE infections was not able to evaluate the effects of intensive care unit stay on
(Table 2) using a univariate logistic regression model. Recent anti- risk of CRE infection as a result of matching patients based on
biotic therapy (P < .0001), recent immunosuppression (P ¼ .003), level of care. The CDC reported that E coli were one of the most
and a Charlson Comorbidity Index score
4 (P ¼ .009) were still common CRE isolated1; however, we did not have any patients
statistically significant after accounting for other factors in the with an index carbapenem-resistant E coli isolated by culture.
multivariate logistic regression model. Recent hospitalization This may be the result of geographic variation of common
(P ¼ .12) became statistically nonsignificant, and transfer from organisms. Finally, these results are specific for NHFMC and
another health care facility (P ¼ .42) remained statistically need to be validated at other institutions before becoming
nonsignificant (Table 3). generalizable.
Recent antibiotic therapy, recent immunosuppression, and In conclusion, this study included all CRE and sources of infec-
Charlson Comorbidity Index score
4 were incorporated into a CRE tion at our facility. Recent antibiotic therapy, immunosuppressive
risk prediction scoring model (Table 4). The diagnostic performance therapy, and Charlson Comorbidity Index score
4 were found to
of various score cutoffs for the CRE risk prediction scoring model is be significant risk factors for developing a CRE infection. A bedside
summarized in Table 5. A score
5 using the CRE risk prediction scoring model was developed including these risk factors. This
model correlated with the highest accuracy (79%), with a corre- model was an excellent predictor of CRE infection and had good
sponding sensitivity of 54% and specificity of 88%. The receiver discriminatory power. A score
5 correlated with the highest ac-
operator characteristic AUC was 0.83 for this model. curacy for predicting CRE infection.
 B.M. Miller, S.W. Johnson / American Journal of Infection Control 44 (2016) 134-7 137

Table 3
Evaluation of risk factors identified

Variable Unadjusted OR (95% CI)* Unadjusted P value* Adjusted OR (95% CI)y Adjusted P valuey
z
Recent antibiotics 16.19 (4.74-55.3) <.0001 13.37 (4.16-61.19) <.0001
Recentx hospitalization 4.97 (1.66-14.89) .001 2.59 (0.79-10.52) .12
Recentz immunosuppression 10.91 (3.25-36.67) <.0001 6.69 (1.85-29.65) .003
Charlson Comorbidity Index score
4 2.8 (1.33-5.88) .009 3.30 (1.34-8.40) .009
Transfer from another health care facility 1.88 (0.92-3.85) .098 1.41 (0.59-3.45) .42

CI, confidence interval; OR, odds ratio.

*Obtained through univariate logistic regression.
y
Obtained through multivariate logistic regression, adjusted for other factors.
z
Within the last 90 days before obtaining index culture.
x
Within the last year before index hospitalization.

Table 4 3. Schwaber MJ, Klarfeld-lidji S, Navon-venezia S, Schwartz D, Leavitt A,

CRE scoring model Carmeli Y. Predictors of carbapenem-resistant Klebsiella pneumoniae acquisi-
tion among hospitalized adults and effect of acquisition on mortality.
Attribute No. of points Antimicrob Agents Chemother 2008;52:1028-33.
4. Lee GC, Lawson KA, Burgess DS. Clinical epidemiology of carbapenem-resistant
Recent* antibiotics 4
enterobacteriaceae in community hospitals: a case-case-control study. Ann
Recent* immunosuppression 3
Pharmacother 2013;47:1115-21.
Charlson Comorbidity Index score
4 2 5. Clinical Laboratory Standards Institute (CLSI). Performance standards for anti-
*Within the last 90 days before obtaining index culture. microbial susceptibility testing: 19th informational supplement. Wayne [PA]:
CLSI; 2009. CLSI document M100eS19.
6. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying
prognostic comorbidity in longitudinal studies: development and validation.
J Chronic Dis 1987;40:373-83.
Table 5 7. Johnson SW, Anderson DJ, May DB, Drew RH. Utility of a clinical risk factor
Distribution of scores with diagnostic cutoff performance scoring model in predicting infection with extended-spectrum b-lactamase-
producing enterobacteriaceae on hospital admission. Infect Control Hosp
Score TP FP TN FN Se Sp PPV NPV Acc
Epidemiol 2013;34:385-92.

0 41 123 0 0 100 0 25 0 25 8. Furuno JP, Harris AD, Wright MO, McGregor JC, Venezia RA, Zhu J, et al. Pre-

2 41 72 51 0 100 41 36 100 56 diction rules to identify patients with methicillin-resistant Staphylococcus

3 39 55 68 2 95 55 41 97 65 aureus and vancomycin-resistant enterococci upon hospital admission. Am J

4 39 54 69 2 95 56 42 97 66 Infect Control 2004;32:436-40.

5 22 15 108 19 54 88 59 85 79 9. Tumbarello M, Trecarichi EM, Bassetti M, De Rosa FG, Spanu T, Di Meco E, et al.

6 21 15 108 20 51 88 58 84 79 Identifying patients harboring extended-spectrum-b-lactamase-producing
Enterobacteriaceae on hospital admission: derivation and validation of a

7 10 3 120 31 24 98 77 79 79
scoring system. Antimicrob Agents Chemother 2011;55:3485-90.

9 5 0 123 36 12 100 100 77 78
10. Gupta N, Limbago BM, Patel JB, Kallen AJ. Carbapenem-resistant
Acc, accuracy; FN, false negatives; FP, false positives; NPV, negative predictive value; enterobacteriaceae: epidemiology and prevention. Clin Infect Dis 2011;
PPV, positive predictive value; Se, sensitivity; Sp, specificity; TN, true negatives; TP, 53:60-7.
true positives. 11. Tacconelli E, Karchmer AW, Yokoe D, D’Agata EM. Preventing the influx of
vancomycin-resistant enterococci into health care institutions, by use of a
simple validated prediction rule. Clin Infect Dis 2004;39:964-70.
12. Tacconelli E, Cataldo MA, De AG, Cauda R. Risk scoring and bloodstream
References infections. Int J Antimicrob Agents 2007;30(Suppl 1):S88-92.
13. Harbarth S, Sax H, Fankhauser-Rodriguez C, Schrenzel J, Agostinho A, Pittet D.
1. Centers for Disease Control and Prevention. Antibiotic Resistance Threats in the Evaluating the probability of previously unknown carriage of MRSA at hospital
United States, 2013. Available at: http://www.cdc.gov/drugresistance/threat- admission. Am J Med 2006;119:275.e15-23.
report-2013/. Accessed August 11, 2014. 14. Van GJ, Florence E, Van den Ende J. Validation of clinical scores for risk
2. Martin ET, Tansek R, Collins V, Hayakawa K, Abreu-Lanfranco O, Chopra T, et al. assessment. Clin Infect Dis 2012;54:1520-1.
The carbapenem-resistant Enterobacteriaceae score: a bedside score to rule out 15. Zweig MH, Campbell G. Receiver-operating characteristic (ROC) plots: a
infection with carbapenem-resistant Enterobacteriaceae among hospitalized fundamental evaluation tool in clinical medicine. Clin Chem 1993;39:
patients. Am J Infect Control 2013;41:180-2. 561-77.