1. Emerging Waves of Carbapenem Resistance among Gram-negative Pathogens at a Tertiary Center
Presenter: Julie Paronish
Faculty Advisor: Nancy W. Glynn, PhD
Internship Preceptor: Ryan K. Shields, PharmD, MS

2. Background
Antibiotic resistance has emerged as a healthcare crisis around the world
In the United States alone, more than $30 billion dollars a year are spent on treating drug-resistant infections
Given the limited therapeutic options for treatment of antibiotic-resistant pathogens, patient morbidity and mortality are disproportionately high
Carbapenems are considered one of the last lines of defense against drug-resistant bacteria; however, rates of carbapenem-resistant organisms (CROs) have increased

3. Background
Despite recent global trends, the epidemiology of CROs at individual centers remains poorly defined
So too are patient or hospital factors associated with the emergence of CROs
At UPMC Presbyterian hospital, carbapenem resistance is now common across many Gram-negative pathogens
Overall burden of CROs, and the pathogen-specific impact on patient outcomes are unknown

4. Objectives & Hypothesis
Define the epidemiology of CROs at UPMC
Identify associations between trends in antibiotic use and CROs
Describe characteristics and outcomes of patients with CROs
Hypotheses
Carbapenem resistance has emerged over time across many pathogens
Increased carbapenem usage is associated with increasing rates of CROs at UPMC
Patient outcomes vary by pathogen, severity of illness, and underlying diseases

5. Methods Microbiology data extracted from 2000 – 2015
Data sources: MISYS and Sunquest
Inpatient locations only – UPMC Presbyterian
Carbapenem resistance was defined by the 2016 CLSI interpretive criteria, and applied retrospectively throughout the study
Unique patients were identified by pathogen for epidemiology analysis (e.g. patients may be re-included for each pathogen)
For outcome analysis, unique patients were identified by the first CRO isolated (e.g. patients were not re-included)
Social Security Death Index to determine time to death
Inpatient was defined by institution antibiogram methodology.

6. Statistical Analysis
Epidemiological trends in CROs and antibiotic daily defined doses were determined by linear regression
Time series, cross-correlation regression analysis was used to compare trends in CROs as a function of antibiotic consumption
Kaplan-meier graphs were used to plot survival over time, and compared by the log-rank test
In univariate analysis to identify predictors of death, continuous and categorical variables were compared by student’s t-test and chi-square, respectively
In multivariate analysis, a logistic regression model was built with stepwise backward selection procedures using variables with a P-value <0.10
Using coefficients from the logistic regression model, a prediction equation was derived by fitting the data to an inverse probability model

7. Gram-negative pathogens at UPMC
97,747 isolates from 42,070 patients were identified
Organism n
% CRO
baumannii
2,954
36.46%
C. freundii
2,209
2.72%
E. coli
28,676
1.14%
E. aerogenes
3,230
5.91%
E. cloacae
6,487
6.81%
K. oxytoca
3,370
2.05%
K. pneumoniae
15,864
9.30%
M. morganii
1,199
4.25%
P. mirabilis
6,372
2.57%
P. aeruginosa
22,851
22.58%
S. marcescens
4,535
4.26%
Organism n
% CRO
baumannii
2,954
36.46%
C. freundii
2,209
2.72%
E. coli
28,676
1.14%
E. aerogenes
3,230
5.91%
E. cloacae
6,487
6.81%
K. oxytoca
3,370
2.05%
K. pneumoniae
15,864
9.30%
M. morganii
1,199
4.25%
P. mirabilis
6,372
2.57%
P. aeruginosa
22,851
22.58%
S. marcescens
4,535
4.26%

8. Targeted CROs (2000 – 2015) 84,597 isolates from 37,823 patients
8,864 isolates from 4,994 patients were defined as CRO
In our tertiary center, Pseudomonas has consistently been the cause of the largest quantity of unique patients over the pas 15 years. KLPN and ACAT both showed increases beginning in 2006, while an increase of ENT in 2009 is noted. *Did any testing methods occur during these years? What else could account for drastic increases.

9. Rates of Carbapenem Resistance by Pathogen
Time series linear regression showed that Carbapenem resistance has increased over time for all organisms (P<.05). The most substantial increase is seen in ACAT beginning in 2006.

10. Incidence of Carbapenem Resistance
Pathogen n
Rate per 1,000 patient days
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
A. baumannii
547
0.04
0.05
0.02
0.07
0.19
0.32
0.36
0.39
0.34
0.2
0.18
E. coli
372
0.03
0.08
0.1
0.14
0.11
0.12
Enterobacter
536
0.01
0.16
0.3
0.38
0.48
K. pneumoniae
650
0.06
0.24
0.26
0.43
0.52
0.33
P. aeruginosa
2267
0.74
0.7
0.64
0.53
0.61
0.68
0.81
0.82
0.88
0.83
0.96
0.95
1.03
1.11
1.09
S. marcescens
127

11. Recurrence rates by pathogen
Note that recurrence for PSAR same pathogen is significantly higher than recurrence of other organisms for PSAR.

12. Antibiotic daily defined doses (DDDs)
P-VALUE
R-SQUARED
0.0000
0.9242
0.0001
0.7005
0.0641
0.2240
Carbapenem DDD’s/1000 patient days increased over the study period (P<.001) ; cephalosporin, piperacillin/tazobactam, and aminoglycoside DDDs were unchanged, while fluoroquinolone DDDs were decreased (P<0.001).

13. Association between CROs and DDDs
Cross-correlation regression analysis showed that increased carbapenem use and the emergence of CROs occurred simultaneously
All CROs
By pathogen
LAG
R-squared
ACAT -3
0.6106
ECOL
0.8868
ENT
0.8534
KLPN
0.8790
PSAR
0.8616
SERM
0.6258
TOTAL
0.9070
By cross-correlation regression analysis, carbapenem DDDs correlated with the emergence of CROs at a zero-lag time interval (R2=0.90, P<0.001), indicating that increased consumption and emergence of CROs occurred simultaneously.

14. Patient characteristics
 Factor
All Patients (n=4,994)
Median Age, years (std dev)
57.8(± 15.9)
Male, no. (%)
2,811(56)
ICU at time of culture, no. (%)
2,064(41)
Solid Organ Transplant Recipient, no. (%)
1,297(26)
CROs were identified from 4,994 unique patients; mean age was 57.8 years (± 15.9), 56% were men, and 41% resided in the intensive care unit (ICU) at the time of isolation. 26% of patients were solid-organ transplant recipients. 30- and 90-day mortality rates were 19% and 32%, respectively. 30d mortality rates were higher for Acat (28%, P<0.001), and lower for Psar (17%, P<0.001), compared to other pathogens.

15. Kaplan-meier survival estimates
By log rank test:
ACAT was associated with a higher rate of 30- and 90- mortality compared to all other pathogens
PSAR was associated with a lower rate of 30- and 90- mortality compared to ACAT and KLPN

16. Mortality Rate by Culture Source
Respiratory
Urine
D. Wound
Blood
S. Wound
Other
ACAT (589)
182 8 30 19 4
ECOL (232) 17 12 14
ENT (487) 71 39 15 2
KLPN (750)
103 33 54 3
PSAR (2794)
600 53
114
SERM (142) 24 7
TOTAL
999
171
235
150 26

17. Logistic Regression Male 2,237 (55.6%) 574 (59.1%) 0.0900 Age > 65
Factor
Alive (n=4,022)
Dead (n=972)
P-value
Male
2,237 (55.6%)
574 (59.1%)
0.0900
Age > 65
226 (5.6%)
406 (41.8%)
<0.0001
Residence in ICU
1,410 (35.1%)
654 (67.3%)
SOT
1,104 (27.4%)
193 (19.9%)
0.0020
Organism
Baumannii
422 (10.5%)
167 (28.4%)
E. coli
195 (4.8%)
37 (15.9%)
0.1936 
Enterobacter
396 (9.8%)
91 (18.7%)
 0.6921
K. Pneumoniae
588 (14.6%)
162 (21.6%)
 0.1204
Pseudomonas
2,311 (57.5%)
483 (17.3%)
 <0.0001
S. marcescens
110 (2.7%)
32 (22.5%)
 0.4063
Culture
Respiratory
2,233 (55.5%)
614 (63.2%)
Urine
661 (16.4%)
89 (9.2%)
D. Wound
72 (18.1%)
143 (14.7%)
Blood
288 (7.2%)
105 (10.8%)
 0.0002
S. Wound
91 (2.3%)
19 (2.0%)
 0.6419

18. Estimating mortality using prediction model
Rationale
Can use coefficients from logistic regression model to make predictions for individual patients
Equation
p= 𝑒 𝑎+𝐵1𝑋1+𝐵2𝑋2+..𝐵𝑝𝑋𝑝 1+ 𝑒 𝑎+𝐵1𝑋1+𝐵2𝑋2+..𝐵𝑝𝑋𝑝
Variable
β-coefficient
Intercept
Age
Male
ICU
SOT
Organism
ACAT
ECOL
ENT
KLPN
PSAR
Culture
RESP
URINE
DWOUN
BLOOD
SWOUN

19. Estimating mortality using prediction model
Example 1
65 year old, non-transplant man in the ICU with A. baumanii in blood cultures
Predicted mortality = 54.87%
Example 2
45 year old, liver transplant female on a medical ward with K. pneumoniae in urine cultures
Predicted mortality = 5.61%
Example 3
79 year old, non-transplant man in the ICU with P. aeruginosa in respiratory cultures
Predicted mortality = 41.88%

20. Conclusions CROs have emerged in waves over time at UPMC
Antibiotic consumption does not appear to be associated with the epidemiology of CROs
CROs are associated with high rates of mortality
Identification of A. baumannii, residence in the ICU, and older age are important predictors of death
Solid-organ transplant recipients, on the other hand, have lower rates of death following isolation of CROs
Active surveillance and infection control strategies will be useful to identify and control the emergence of other CROs, respectively