Академический Документы
Профессиональный Документы
Культура Документы
S0195-6701(16)00054-2
DOI:
10.1016/j.jhin.2016.01.007
Reference:
YJHIN 4726
To appear in:
Please cite this article as: Kengkla K, Charoensuk N, Chaichana M, Puangjan S, Rattanapornsompong
T, Choorassamee J, Wilairat P, Saokaew S, Clinical risk scoring system for predicting extendedspectrum -lactamase-producing Escherichia coli infection in hospitalized patients, Journal of Hospital
Infection (2016), doi: 10.1016/j.jhin.2016.01.007.
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
K. Kengkla et al.
Clinical risk scoring system for predicting extended-spectrum -lactamase-producing
Escherichia coli infection in hospitalized patients
K. Kengklaa,b, N. Charoensuka, M. Chaichanaa, S. Puangjana, T. Rattanapornsomponga, J.
a
RI
PT
Corresponding author. Address: Center of Health Outcomes Research and Therapeutic Safety
SC
M
AN
U
TE
D
Methods: The study retrospectively collected eligible patients with a positive culture for E. coli
during 2011 to 2014. The risk scoring system was developed using variables independently
associated with ESBL-EC infection through logistic regression-based prediction. Area under the
of the model.
EP
receiveroperator characteristic curve (AuROC) was determined to confirm the prediction power
Findings: Predictors for ESBL-EC infection were male gender [odds ratio (OR): 1.53], age 55
AC
C
years (OR: 1.50), healthcare-associated infection (OR: 3.21), hospital-acquired infection (OR:
2.28), sepsis (OR: 1.79), prolonged hospitalization (OR: 1.88), history of ESBL infection within
one year (OR: 7.88), prior use of broad-spectrum cephalosporins within three months (OR:
12.92), and prior use of other antibiotics within three months (OR: 2.14). Points scored ranged
from 0 to 47, and were divided into three groups based on diagnostic performance parameters:
low risk (score: 08; 44.57%), moderate risk (score: 911; 21.85%) and high risk (score: 12;
33.58%). The model displayed moderate power of prediction (AuROC: 0.773; 95% confidence
interval: 0.7420.805) and good calibration (HosmerLemeshow 2 = 13.29; P = 0.065).
1
ACCEPTED MANUSCRIPT
Conclusion: This tool may optimize the prescribing of empirical antibiotic therapy, minimize
time to identify patients, and prevent spreading of ESBL-EC. Prior to adoption into routine
clinical practice, further validation study of the tool is needed.
Keywords:
RI
PT
SC
M
AN
U
Escherichia coli (ESBL-EC), which has now spread across all regions of the world, including
Thailand.13 These infections increase morbidity, mortality, length of stay, and treatment-related
costs in seriously infected patients.46
TE
D
over-use of carbapenems for empirical treatment of suspected ESBL-EC infection may promote
carbapenem-resistant bacteria such as E. coli, Acinetobacter baumannii, and Pseudomonas
aeruginosa, as well as adding to antibiotic costs.7,9 This needs to be balanced against the risks of
delay in initiating appropriate antibiotic therapy in hospitalized cases of infection with ESBL-EC,
EP
AC
C
empirical carbapenem therapy. Thus, treatment failure, infectious complications, antibiotic costs,
and the risk of selecting carbapenem-resistant bacteria should decline.1214 A risk discrimination
tool that can identify patients likely to have ESBL-EC might therefore assist with rational
antibiotic prescribing and the early implementation of infection control precautions.
A previously published clinical prediction scoring tool was reported to display good
discrimination and excellent predictive value for both community- and hospital-acquired
infections with ESBL-producing Enterobacteriaceae.15,16 When we applied these scoring systems
on our patients, only two factors predicting possession of ESBL-EC remained: previous
hospitalization and recent antibiotic therapy with -lactam/-lactamase inhibitors. Both the model
ACCEPTED MANUSCRIPT
of Tumbarello et al. and the Duke Model had poor predictive values, with area under the
receiveroperator characteristic curve (AuROC) of 0.530 and 0.548, respectively.15,16 This may
be because the epidemiology and phenotypic determinants of different bacteria vary between
geographical areas, and even between institutions in the same geographical area.17,18
RI
PT
Therefore, this study was intended to develop a reliable, easy-to-use risk scoring system
to predict ESBL-EC infection based on local epidemiology in Thailand, with the aim of
improving carbapenem antibiotic stewardship.
Methods
Patients
SC
Phayao hospital is a 400-bed secondary care medical centre located in the north of
Thailand. All patients who had positive cultures for E. coli in any specimen, and who were
M
AN
U
admitted to hospital from January 2011 to December 2014, were eligible for inclusion. Patients
readmitted to hospital within 30 days were excluded, but those readmitted after more than 30
days were included. Patients with missing data on the electronic database were excluded. Patients
were classified into two groups: ESBL-EC (case) and non-ESBL-EC (control).
Definitions
Hospital-acquired infection (HAI) was defined as an infection that occurred >48 h after
TE
D
EP
or more days within 90 days before onset of the infection; (iv) resident in a nursing home or longterm care facility.20 Community-acquired infection (CAI) was defined as an infection that was
AC
C
present on admission to hospital, or that occurred within 48 h of admission, where the criteria for
HCAI were not fulfilled.19
manifested by two or more of temperature >38 or <36C, heart rate >90 bpm, respiratory rate
>20 bpm or PaCO2 <32 mmHg, and white blood cell count >12,000 or <4000 cells/mm3 in a
patient with suspected infection.21,22
Predictor parameters
Clinical characteristics, medical histories, and drug exposures were obtained from the
computerized medical database. Parameters of interest included age, gender, types of infection
(HAI, HCAI, or CAI), hospital location at time of infection (general ward, surgical ward, and
ACCEPTED MANUSCRIPT
intensive care unit), sepsis, previous history of E. coli infection, history of ESBL-EC infection
within the preceding year, hospital stays before infection, interventions (recent surgery,
percutaneous drainage, peritoneal dialysis, Foley catheter, mechanical ventilator, and nasogastric
tube), comorbidities (diabetes mellitus, cardiovascular diseases, neurological diseases, pulmonary
RI
PT
diseases, hepatobiliary diseases, chronic kidney disease, acute kidney injury, solid tumour cancer,
and autoimmune diseases), medication history (corticosteroids and other immunosuppressive
agents), and antibiotic exposures in the preceding three months.
Data analysis
SC
range), mean (SD), or frequency count (percentage), as appropriate. The associations between
candidate risk factors and ESBL infection were analysed using logistic regression with variance-
M
AN
U
corrected model and summarized with odds ratios (ORs) and 95% confidence intervals (CIs).
Each factor was screened for an association with ESBL infection via univariate logistic
regression; those with 0.1 were carried forward in multivariate analyses. Candidate risk
factors were then plugged into a multivariate logistic regression using a cluster-robust variancecorrected model with backward elimination. We removed variables with P > 0.05. Odds ratio and
95% CI were calculated to evaluate the strength of any association that emerged. Any variable
TE
D
with P < 0.05 was retained in the final model. A points-based system was developed by
coefficient from the final regression model. The coefficient regression of each variable was then
divided by the smallest coefficient number and rounded to the nearest integer. The prediction
model performed discriminatory power by AuROC and the calibration efficiency was tested
EP
using the HosmerLemeshow test. Potential interactions were also checked in the predictive
model. Collected data were processed using Stata, version 12.1 (StataCorp., College Station, TX,
AC
C
USA).
This study was approved by the Research Ethics Committee of the Division of Research
Administration and Educational Quality Assurance, University of Phayao, Phayao, Thailand (No.
5702010001).
Results
Of the 3159 patients with positive cultures for E. coli assessed for eligibility, 810 were
finally included in the study: 443 patients in the ESBL-EC group (cases) and 367the patients in
non-ESBL-EC group (controls) (Figure 1). The majority of patients were female 456 (56.30%)
and mean age (SD) was 63.32 (19.12) years. There were 446 (55.0%) urinary tract infections, 100
ACCEPTED MANUSCRIPT
(12.3%) lower respiratory infections, 85 (10.5%) abdominal infections, 56 (6.9%) skin infections,
and 20 (2.5%) bloodstream infections.
Patients with ESBL-EC and non-ESBL-EC infection are summarized by univariate
analysis in Table I. The patients with ESBL-EC infection were older, a higher proportion was
RI
PT
male, and there were significant associations with hepatic insufficiency, presence of sepsis, recent
ESBL infection within one year, recent surgery, mechanical ventilator, prolonged hospitalization
(>7 days), and prior use of broad spectrum cephalosporins and fluoroquinolones.
Significant predictors
Significant (P < 0.05) multivariate predictors for ESBL-EC infection were male sex (OR:
SC
1.53; 95% CI: 1.112.12), age 55 years (1.50; 1.042.16), HCAI (3.21; 2.114.88), HAI (2.28;
1.483.50), sepsis (1.79; 1.292.49), prolonged hospitalization (1.88; 1.093.23), history of
M
AN
U
ESBL-EC infection within one year (7.88; 2.2427.68), prior use of broad-spectrum
cephalosporins (12.92; 4.4237.78), and prior use of other antibiotics (2.14; 1.443.20).
Scoring system
The scores were divided by significant coefficient predictors with the smallest coefficient
(0.40) then rounded to the nearest integer. The scores ranged from 0 to 13. Each score was
summed to overall ESBL infection risk scores, which could range from 0 to 47 (Table II).
TE
D
Discrimination
The average ESBL infection risk scores in case and control groups were 12.85 7.02 and
6.86 4.38, respectively (P < 0.001) (Figure 2, Table III).
The scores discriminated patients with ESBL-EC infection from those with non-ESBL-
EP
EC infection with moderate-level validity (AuROC 0.773, 95% CI 0.7420.805) in a wellcalibrated predictive model (HosmerLemeshow 2 = 13.29; P = 0.065).
AC
C
Clinical predictions
The risk scoring system was developed as a clinical prediction rule by determining the
cut-off according to the discrimination plot and performance of diagnostic parameters in order to
identify patients at probability risk of ESBL-EC infection (Table IV).
We considered and selected two cut-off points. A cut-off score of 9 had good sensitivity
(74%) and moderate specificity (66%); positive predictive value (PPV) and negative predictive
value (NPV) were 73% and 68%, respectively, and accuracy (70%) was acceptable. With a cutoff score of 12, the sensitivity declined (53%), but specificity rose considerably (90%). NPV
and PPV were 61% and 86%, respectively, and overall accuracy (69%) was comparable with that
of the lower cut-off value (Table IV).
ACCEPTED MANUSCRIPT
ESBL infection scores were classified into three risk levels. Low risk level was defined as
overall scores of 8 points (N = 361, 44.57%), moderate risk as scores of 911 points (N = 177,
21.85%), and high risk as scores of 12 points (N = 272, 33.58%). Using this scoring system, 244
of 810 (30.12%) cases with E. coli were correctly classified as low risk (score 08), with 117
RI
PT
(14.44%) cases misclassified (risk underestimated). At the moderate risk level (score 911), the
score correctly identified 93 of 810 (11.48%) cases; however, 84 (10.38%) moderate risk level
cases did not have ESBL-EC. At the high risk levels (score 12), the score correctly classified
233 of the 810 (28.77%) cases; only 39 (4.81%) cases were misclassified (risk overestimated)
SC
(Table III). Overall, 70.37% of patients were correctly classified as being at risk or not at risk of
M
AN
U
In this study, we established the first clinical risk scoring system for predicting ESBL-EC
infection in Asia using clinical and demographic variables at time of infection. Patients were
classified into one of three groups of likelihood of having ESBL-EC (Figure 3) with acceptable
performance (AuROC: 0.773).23 The tool may therefore be clinically useful in clinical decisionmaking for hospitalized patients. Whereas predictive scores have been developed in previous
research for predicting ESBL-EKP infection in the community or in hospital , our study included
TE
D
Our predictive model consists of the nine predictors of ESBL-EC infection. At least three
of these (age >55 years, and exposure to broad-spectrum cephalosporins and other antibiotics
within three months) are consistent with predictors in previous models.15,16 Additionally, we
EP
found that male gender, HCAI, HAI, prolonged hospitalization, sepsis, and ESBL-EC infection
within one year were new predictors of ESBL-EC. However, in contrast to Charlsons
AC
C
comorbidity index, recent urinary catheterization and immunosuppression were not predictors in
our study.15,16 We also noted that Charlsons comorbidity index score was not simple and
convenient to use in clinical practice.
Both c-statistics of the model and goodness-of-fit by the HosmerLemeshow 2-statistic
confirmed a good discriminatory power of the AuROC, although the overall performance of our
model was poorer than in both previous predictive models.15,16 However, our model appears to be
the best available for identifying patients at risk of having infection with ESBL-EC.24
Using our model, if the intention were to screen hospitalized patients to determine the
likelihood of infection with ESBL-EC, a cut-off point with high sensitivity and low specificity
could be employed. On the other hand, if application were intended to guide therapeutic
ACCEPTED MANUSCRIPT
intervention (in particular to avoid overuse of carbapenems), then a cut-off model point with a
lower sensitivity and higher specificity would be appropriate.25 In the model of Tumbarello et al.
the cut-off of at least 3 points had a very high sensitivity (93%) and NPV (97%), but very low
specificity (62%) and PPV (45%).15 By raising the cut-off to 6 points, sensitivity and NPV fell to
RI
PT
63% and 88%, but specificity and PPV increased to 95% and 80%. In the Duke model, with a
cut-off of 8 points, good specificity and PPV (95% and 79%, respectively) contrasted with
poorer sensitivity (58%).16
In our study, the cut-off was based on assessment of sensitivity, specificity, PPV, NPV
and accuracy. The best cut-off points for predicting any risk of ESBL-EC infection was 9
SC
points, which had an accuracy of 70% and acceptable underestimation at about 14%. However,
this cut-off had moderate sensitivity (74%) and NPV (68%) and inadequate specificity, including
M
AN
U
overestimation at 15%. For this reason it is not the best cut-off point for predicting high risk of
ESBL-EC infection, nor for making empiric antibiotic prescribing decisions. Here, a cut-off point
with a higher specificity and PPV is required. We used a cut-off score for separating high from
moderate risk of ESBL-EC infection of 12 points. Specificity and PPV are increased to 89% and
86%, but sensitivity is lower at 53%.
We propose that cut-off scores may be used to classify patients into three groups with
TE
D
respect to empiric antibiotic therapy decisions. Patients with score of 8 are low risk, and do not
require empiric therapy to cover ESBL-EC. Patients with score of 911 are moderate risk (RR:
1.73; 95% CI: 1.362.20; P < 0.001]. We recommend the use of agents such -lactam/lactamase inhibitors or fluoroquinolones as empiric therapy for this group. Finally, patients with
EP
score 12 are defined as high risk (RR: 4.83; 95% CI: 3.586.53; P < 0.001], and for these we
recommend empiric therapy with a carbapenem.
AC
C
Our study has important limitations. First, we collected data only on patients with E. coli
infection, and we do not know whether the predictive score applies to other ESBL-producing
Enterobacteriaceae (e.g. Klebsiella pneumonia, Proteus mirabilis). Second, we collected data
from electronic medical charts retrospectively; it is likely that some of these data were
incomplete or inaccurate, especially information on previous admissions to other hospitals and
antibiotic histories. We recommended conducting a prospective study, reviewing fuller data sets,
and interviewing patients instead. Third, the predictive score was not validated due to an
inadequate sample size; full validation would be required before adoption of a scoring system
into clinical practice. Finally, our study was a single-centre retrospective study, and the findings
may not be generalizable, even within Thailand or South East Asia.
ACCEPTED MANUSCRIPT
In conclusion, a novel scoring system was developed for predicting ESBL-EC infection
that could certainly identify patients with these infections. The scores depended on nine
predictors, most of which are easy and convenient to ascertain; however, we do acknowledge that
accurate information about hospitalization in other hospitals and on previous antibiotic therapy
RI
PT
may be difficult to obtain. Nevertheless, we believe that predictive tools such as ours may
improve the effectiveness of medical care, by allowing early targeting of interventions for
developed, validated, and introduced into clinical practice as a key part of our fight against MDR
SC
Gram-negative bacteria.
Acknowledgements
M
AN
U
The authors wish to thank the authorities of the Phayao hospitals for their support. We
thank Prof. G.M. Oderda for reviewing the manuscript before submission.
Conflict of interest statement
None declared.
Funding sources
The study was partially funded by a grant from School of Pharmaceutical Sciences, University
References
1.
TE
D
2.
EP
3.
AC
C
Giske CG, Monnet DL, Cars O, Carmeli Y. Clinical and economic impact of common
multidrug-resistant gram-negative bacilli. Antimicrob Agents Chemother 2008;52:813821.
5.
Maslikowska JA, Walker SA, Elligsen M, et al. Impact of infection with extendedspectrum beta-lactamase-producing Escherichia coli or Klebsiella species on outcome and
hospitalization costs. J Hosp Infect 2016;92:3341.
ACCEPTED MANUSCRIPT
6.
RI
PT
7.
9.
10.
SC
Dautzenberg MJ, Wekesa AN, Gniadkowski M, et al. The association between colonization
M
AN
U
Schwaber MJ, Carmeli Y. Mortality and delay in effective therapy associated with
extended-spectrum beta-lactamase production in Enterobacteriaceae bacteraemia: a
systematic review and meta-analysis. J Antimicrob Chemother 2007;60:913920.
12.
13.
TE
D
EP
14.
15.
AC
C
2013;84:294299.
Tumbarello M, Trecarichi EM, Bassetti M, et al. Identifying patients harboring extendedspectrum-beta-lactamase-producing Enterobacteriaceae on hospital admission: derivation
and validation of a scoring system. Antimicrob Agents Chemother 2011;55:34853490.
16.
Johnson SW, Anderson DJ, May DB, Drew RH. Utility of a clinical risk factor scoring
model in predicting infection with extended-spectrum beta-lactamase-producing
Enterobacteriaceae on hospital admission. Infect Control Hosp Epidemiol
2013;34:385392.
17.
Drawz SM, Porter S, Kuskowski MA, et al. Variation in resistance traits, phylogenetic
backgrounds, and virulence genotypes among Escherichia coli clinical isolates from
ACCEPTED MANUSCRIPT
adjacent hospital campuses serving distinct patient populations. Antimicrob Agents
Chemother 2015;59:53315339.
18.
19.
RI
PT
2010;70:313333.
Siegman-Igra Y, Fourer B, Orni-Wasserlauf R, et al. Reappraisal of community-acquired
bacteremia: a proposal of a new classification for the spectrum of acquisition of bacteremia.
Clin Infect Dis 2002;34:14311439.
20.
21.
SC
infection: a systematic review 10 years after the first proposal. BMC Med 2014;12:40.
Bone RC, Balk RA, Cerra FB, et al. Definitions for sepsis and organ failure and guidelines
M
AN
U
for the use of innovative therapies in sepsis. The ACCP/SCCM Consensus Conference
Committee. American College of Chest Physicians/Society of Critical Care Medicine.
Chest 1992;101:16441655.
22.
Dellinger RP, Levy MM, Rhodes A, et al. Surviving sepsis campaign: international
guidelines for management of severe sepsis and septic shock: 2012. Crit Care Med
2013;41:580637.
Steyerberg EW, Vickers AJ, Cook NR, et al. Assessing the performance of prediction
TE
D
23.
reply 2220.
van Griensven J, Florence E, Van den Ende J. Validation of clinical scores for risk
AC
C
25.
EP
ACCEPTED MANUSCRIPT
Table I
Distribution of candidate variable for extended-spectrum -lactamase (ESBL) Escherichia coli infection
ESBL
Non-ESBL
E. coli
E. coli
(N = 443)
(N = 367)
64.35 (17.30)
95% CI
62.28 (20.94)
1.01
0.991.01
0.134
344 (77.65%)
258 (70.30%)
1.47
1.062.04
0.023
213 (48.08%)
141 (38.42%)
1.48
1.102.00
Community-acquired
131 (29.57%)
233 (63.49%)
1.00
Reference
Healthcare-associated
179 (40.41%)
77 (20.98%)
4.13
2.925.85
<0.001
Hospital-acquired
133 (30.02%)
57 (15.53%)
4.15
2.875.99
<0.001
General ward
298 (67.27%)
280 (76.29%)
1.00
Reference
26 (5.87%)
11 (3.00%)
2.22
1.074.61
0.032
Surgical ward
119 (26.86%)
76 (20.71%)
1.47
1.042.08
0.029
87 (19.64%)
28 (7.63%)
2.96
1.884.67
<0.001
Presence of sepsis
292 (65.91%)
203 (55.31%)
1.56
1.172.09
0.003
50 (11.29%)
11.55
3.9433.80
<0.001
55 years
Male sex
4 (1.09%)
AC
C
Intervention history
TE
D
0.010
M
AN
U
Types of infection
SC
P-value
RI
PT
OR
EP
Characteristics
Recent surgery
132 (29.80%)
72 (19.62%)
1.74
1.242.43
0.001
Foley catheter
196 (44.24%)
148 (40.33%)
1.17
0.881.57
0.281
Mechanical ventilator
77 (17.38%)
44 (11.99%)
1.54
1.032.31
0.034
101 (22.80%)
67 (18.26%)
1.32
0.921.91
0.136
Cardiovascular diseases
66 (14.90%)
45 (12.26%)
1.25
0.821.91
0.292
11
ACCEPTED MANUSCRIPT
OR
95% CI
E. coli
E. coli
(N = 443)
(N = 367)
Neurological diseases
52 (11.74%)
Pulmonary diseases
38 (10.35%)
1.15
0.731.81
0.540
61 (13.77%)
39 (10.63%)
1.34
0.842.14
0.217
Hepatic insufficiency
22 (4.97%)
31 (8.45%)
0.57
0.321.01
0.055
83 (18.74%)
52 (14.17%)
1.40
0.952.06
0.091
29 (6.55%)
21 (5.72%)
1.15
0.642.09
Immunosuppressive therapy
4 (0.90%)
8 (2.18%)
0.41
0.121.44
P-value
RI
PT
Non-ESBL
SC
ESBL
0.637
0.164
M
AN
U
Characteristics
39 (8.80%)
25 (6.81%)
1.32
0.772.25
0.308
13 (2.93%)
9 (2.45%)
1.20
0.502.90
0.681
Broad-spectrum cephalosporins
76 (17.16%)
16 (4.36%)
4.54
2.597.96
<0.001
-Lactam/-lactamase inhibitors
30 (6.77%)
14 (3.81%)
1.83
0.943.57
0.076
Carbapenems
8 (1.81%)
2 (0.54%)
3.36
0.6816.58
0.137
Aminoglycosides
5 (1.13%)
1 (0.27%)
4.18
0.4538.47
0.207
Fluoroquinolones
66 (14.90%)
34 (9.26%)
1.71
1.112.65
0.016
Co-trimoxazole
5 (1.13%)
5 (1.36%)
0.83
0.242.88
0.765
EP
TE
D
Penicillins
AC
C
Odds ratio from univariate logistic regression analysis with variance corrected for 728 clusters in patient number.
ACCEPTED MANUSCRIPT
Table II
Significant predictors and assigned item score
Types of infection
Presence of sepsis
a
Prolonged hospitalization
1.53
Female
Reference
55 years
1.50
<55 years
Reference
Healthcare
Coefficient
Score
1.112.12
0.010
1.042.16
0.030
3.21
2.114.88
<0.001
1.17
Hospital
2.28
1.483.50
<0.001
0.82
Community
Reference
Yes
1.79
No
Reference
Yes
1.88
No
Reference
Yes
7.88
No
c
P-value
0.40
1.292.49
0.001
0.58
1.093.23
0.023
0.63
2.2427.68
0.001
2.06
10
12.92
4.4237.78
<0.001
2.56
13
2.14
1.443.20
<0.001
0.76
Reference
BCEP
EP
Otherd
No
0.43
SC
Age
Male
95% CI
M
AN
U
Gender
OR
RI
PT
Category
TE
D
Predictors
Reference
AC
C
OR, odds ratio; CI, confidence interval; ESBL, extended-spectrum -lactamase; BCEP, broad-spectrum cephalosporin.
Coefficients and odds ratio from multivariate logistic regression analysis with variance corrected for 728 clusters in patient number.
a
b
c
ACCEPTED MANUSCRIPT
Table III
Classification of cases versus controls into low, moderate, and high probability categories and risk-estimation validity
Cases
Control
range
(N = 443)
(N = 367)
Mean SD
12.85 7.02
6.86 4.38
Median (IQR)
6 (410)
12 (816)
Risk-estimation validitya
Over
Correct
Under
(%)
(%)
(%)
08
117 (26.41%)
244 (66.49%)
30.12
14.44
Moderate
911
93 (20.99%)
84 (22.89%)
10.38
11.48
High
12
233 (52.60%)
39 (10.63%)
4.81
28.77
Total
15.19
70.37
14.44
EP
TE
D
AC
C
M
AN
U
Low
RI
PT
Score
SC
Categories
ACCEPTED MANUSCRIPT
Table IV
FP
TN
FN
Se (%)
Sp (%)
PPV (%)
NPV (%)
Acc (%)
442
359
99.77
2.18
55.18
88.89
55.56
442
359
99.77
2.18
55.18
88.89
55.56
421
299
68
22
95.03
18.53
58.47
75.56
60.37
415
277
90
28
93.68
24.52
59.97
76.27
62.35
407
255
112
36
91.87
30.52
61.48
75.68
64.07
366
187
180
77
82.62
49.05
66.18
70.04
67.41
358
173
194
85
80.81
52.86
67.42
69.53
68.15
341
142
225
102
76.98
61.31
70.60
68.81
69.88
326
123
244
117
73.59
66.49
72.61
67.59
70.37
10
300
101
266
143
67.72
72.48
74.81
65.04
69.88
11
278
85
282
165
62.75
76.84
76.58
63.09
69.14
12
233
39
328
210
52.60
89.37
85.66
60.97
69.26
13
211
30
337
232
47.63
91.83
87.55
59.23
67.65
14
173
18
349
270
39.05
95.10
90.58
56.38
64.44
15
146
12
355
297
32.96
96.73
92.41
54.45
61.85
16
117
359
326
26.41
97.82
93.60
52.41
58.77
17
98
360
345
22.12
98.09
93.33
51.06
56.54
SC
TP
AC
C
EP
TE
D
M
AN
U
Score
RI
PT
TP, true positive; FP, false positive; FN, false negative; TN, true negative; Se, sensitivity; Sp, specificity; PPV, positive predictive value; NPV, negative predictive
value; Acc, accuracy.
ACCEPTED MANUSCRIPT
Figure 1. Selection of study patients.
Figure 2. Discrimination of extended-spectrum -lactamase (ESBL) Escherichia coli infection
based on ESBL infection risk scores. Dark grey bars: cases (N = 443); light grey bars: controls (N
= 367).
RI
PT
Figure 3. Distribution of extended-spectrum -lactamase infection risk level. Dark grey bars:
AC
C
EP
TE
D
M
AN
U
SC
16
ACCEPTED MANUSCRIPT
Patients with E.coli infection in Phayao hospital from January 2011
to December 2014 were enrolled (n=3,159)
Excluded
- 277 unavailable data
in electronic database
- 213 duplicate isolate
Study patients
443 cases included for analysis
RI
PT
Control (n=2,226)
Patients with non-ESBL producing E.coli
-ESBL-E.coli
Case (n=933)
Patients with ESBL producing E.coli
Excluded
- 1,658 unavailable data
in electronic database
- 201 repeated isolate
SC
Study patients
367 controls included for analysis
AC
C
EP
TE
D
M
AN
U
M
AN
U
SC
RI
PT
ACCEPTED MANUSCRIPT
AC
C
EP
TE
D
100
90
80
70
60
50
40
30
20
10
0
non-ESBL infection
low
moderate
Risk level
high
AC
C
EP
TE
D
M
AN
U
RI
PT
ESBL infection
SC
ACCEPTED MANUSCRIPT