A Systematic Review and Meta-Analyses Show that Carbapenem Use and Medical Devices Are the Leading Risk Factors for Carbapenem-Resistant Pseudomonas aeruginosa

Abstract

A systematic review and meta-analyses were performed to identify the risk factors associated with carbapenem-resistant Pseudomonas aeruginosa and to identify sources and reservoirs for the pathogen. A systematic search of PubMed and Embase databases from 1 January 1987 until 27 January 2012 identified 1,662 articles, 53 of which were included in a systematic review and 38 in a random-effects meta-analysis study. The use of carbapenem, use of fluoroquinolones, use of vancomycin, use of other antibiotics, having medical devices, intensive care unit (ICU) admission, having underlying diseases, patient characteristics, and length of hospital stay were significant risk factors in multivariate analyses. The meta-analyses showed that carbapenem use (odds ratio [OR] = 7.09; 95% confidence interval [CI] = 5.43 to 9.25) and medical devices (OR = 5.11; 95% CI = 3.55 to 7.37) generated the highest pooled estimates. Cumulative meta-analyses showed that the pooled estimate of carbapenem use was stable and that the pooled estimate of the risk factor “having medical devices” increased with time. We conclude that our results highlight the importance of antibiotic stewardship and the thoughtful use of medical devices in helping prevent outbreaks of carbapenem-resistant P. aeruginosa.

INTRODUCTION

Pseudomonas aeruginosa is one of the most common nosocomial pathogens (1). P. aeruginosa can cause infections in patients with serious underlying disorders, such as a suppressed immune system or cystic fibrosis (CF), or in patients in intensive care units (ICU) (2, 3). Further, infections with P. aeruginosa in such patients lead to increased morbidity and mortality (2–4).

P. aeruginosa is intrinsically resistant to various antibiotics and is capable of acquiring additional resistance by either chromosomal mutations or horizontal gene transfer (5). The most important mechanisms are loss or alteration of outer membrane porins and increased efflux pump activity (6–8). The emergence of multidrug-resistant (MDR) P. aeruginosa is a problem of global concern, and there are currently reports of hospital outbreaks of MDR P. aeruginosa from countries around the world, including the Netherlands (9–13). These outbreaks are frequently caused by Pseudomonas aeruginosa clones with metallo-β-lactamases, such as Verona integron-encoded metallo-β-lactamase (VIM) and imipenemase (IMP). Importantly, outbreaks may be large and sustained, despite the adoption of infection control measures (12, 14).

In 2006, a summary on this subject was published by Falagas and Kopterides, who published a systematic review of the problem (15). However, there have been many more published reports regarding nosocomial (MDR) P. aeruginosa since 2006. Therefore, in the current publication, a more extensive and up-to-date systematic review was performed, focusing on carbapenem resistance and non-CF patients and including conventional and cumulative meta-analyses. The aim of the analysis was to answer the following two questions. First, what are the risk factors for the presence of carbapenem-resistant P. aeruginosa among hospitalized patients? Second, what environmental sources and/or reservoirs were identified in these outbreaks? This knowledge will be useful for worldwide health care centers that are facing the threat of MDR P. aeruginosa and will help in designing strategies to stop the emergence of spread of these MDR pathogens.

MATERIALS AND METHODS

The systematic review and meta-analyses presented in this publication include all of the items in the checklist detailed in the PRISMA guideline (16).

Study and data collection.

Eligible articles were identified by searching PubMed (Medline) and Embase databases. Additional articles were identified by hand searching the reference lists of included reviews. Searches were performed for the period from 1 January 1987 until 27 January 2012. Search terms included “Pseudomonas” as a title word, in combination with the keywords “resistant,” “multidrug resistance,” “VIM,” “IMP,” “metallo-beta-lactamase” or “MBL” and “risk factors,” “determinants,” “outbreak,” “transmission,” “nosocomial,” “health care related,” “health care associated,” “epidemiology,” or “source,” including all possible ways of writing. The authors included peer-reviewed articles relating to carbapenem-resistant P. aeruginosa that also described the risk factors associated with the presence of carbapenem-resistant P. aeruginosa using a multivariate model and in which a nosocomial infection was described. We excluded studies relating to nonhuman infections, studies that included only patients with CF, reviews, commentaries, editorials, letters, and abstracts. We also excluded studies published before 1987, the year of the U.S. approval of imipenem (17). Environmental sources and reservoirs were searched for in both included and excluded studies. A study was excluded from the meta-analyses (i) if it reported only hazard ratios, (ii) if it reported only prevalence ratios or risk ratios, (iii) when confidence intervals were missing, and (iv) if it included only patients with P. aeruginosa bacteremia.

We extracted detailed information from the included studies. We based the classification of studies regarding the different study designs on the description of the methods in a particular study, not on the study design claimed to be used by the authors (e.g., a reported retrospective cohort study can methodologically be a case-control study). We contacted the corresponding and/or first authors of 47 articles by e-mail in order to retrieve the full-text articles or to retrieve missing information.

Study quality.

To assess the quality, risk of bias, and generalizability of the included studies, a quality assessment was performed using the STROBE guidelines for included cross-sectional studies as well as the Newcastle-Ottawa quality assessment scale for included case-control and cohort studies (18, 19). The quality of the studies was not considered an exclusion criterion.

Statistical analysis.

We merged all reported risk factors with a reported odds ratio (OR) and 95% confidence interval (95% CI) into 10 different groups: group 1, carbapenem use; group 2, quinolone use; group 3, vancomycin use; group 4, other antibiotic use; group 5, medical devices; group 6, ICU admission; group 7, underlying diseases; group 8, patient characteristics; group 9, length of hospital stay; and group 10, other. We selected the 10 groups using the results of the systematic review. For each of the first nine groups, a meta-analysis was performed. That was not possible for group 10 (other), as the risk factors were too diverse. An additional meta-analysis was performed for the risk factors quinolone use, vancomycin use, and other antibiotic use together. All meta-analyses were performed using StatsDirect statistical software (Altrinchem, United Kingdom). The risk factors reported by the studies included in the analyses were diverse; therefore, a random-effects model was fitted to the data based on the method of DerSimonian and Laird (20). A P value of <0.05 was considered statistically significant, and no correction was made for multiple testing. The risk of publication bias across the studies was assessed by the Egger and Begg-Mazumdar (Kendall's tau) indicators. Both bias indicators had to show a significant result before it was concluded that publication bias was present. Additionally, two cumulative meta-analyses were performed for groups 1 (carbapenem use) and 2 (medical devices), as these two groups showed highly significant results using conventional meta-analyses. A random-effects model, based on the method of DerSimonian and Laird, was also fitted to these cumulative meta-analyses (20).

RESULTS

Description of included studies.

A total of 1,662 articles were identified when the search results of PubMed and Embase were combined (Fig. 1). After applying exclusion criteria as described in Materials and Methods, 256 articles were read in their entirety (full text) (Fig. 1). The corresponding and/or first authors of 47 of 256 articles were contacted by e-mail. Authors from 19 of 47 articles responded to e-mail requests. Nine full-text articles were received by mail or e-mail, and from four articles missing information was retrieved. For two articles, the requested information was not available. Fifty-three studies were finally included in the analyses after exclusion of articles that did not meet our inclusion criteria as described in Materials and Methods (Fig. 1). These studies represented 3,966 patient cases (ranging from 5 to 345 cases per publication) from 15 different countries (Tables 1 and 2). Eight of 53 of the studies included patients with bacteremia only and are shown in Table 2. All 53 studies had an observational study design and were written in English. Eight studies reported that a retrospective cohort study was performed, whereas conceptually they could be considered case-control studies. Five multicenter studies were also included. The percentage of male gender ranged from 38.6% to 84.0%. Patient age ranged from several days old (neonates) to very old, with 97 years as oldest.

FIG 1.

Flow diagram of study selection for the systematic review on carbapenem-resistant P. aeruginosa.

TABLE 1.

Sources and characteristics of included studies (n = 45) and risk factors for transmission and acquisition of carbapenem-resistant P. aeruginosa based on multivariate analyses^a

Risk factor	No. offactors	Sourcesb	No. of cases	Case-control studies
			Range	No.	OR range
Carbapenem use	19	Harris, 2011 (33); Lautenbach, 2010 (34); Lepelletier, 2010 (35); Cezario, 2009 (36); Mueller, 2008 (37); Onguru, 2008 (27); Pena, 2007 (38); Mentzelopoulos, 2007 (39); Fortaleza, 2006 (25); Ohmagari, 2005 (40); Ozkurt, 2005 (41); 2× Zavascki, 2005 (42); Cao, 2004 (43); Harris, 2002 (26); Troillet, 1997 (44); Carmeli, 1999 (45); Lodise, Jr., 2007 (46); Montero, 2010 (47)	5–354	12	3.6–76.0
Quinolone use	11	van der Bij, 2011 (12); Kohlenberg, 2010 (48); Pena, 2009 (49); Yang, 2009 (50); Pena, 2007 (38); Lautenbach, 2006 (51); Zavascki, 2006 (52); Nouer, 2005 (53); Defez, 2004 (54); Lodise, Jr., 2007 (46); Montero, 2010 (47)	15–354	5	2.5–48.4
Vancomycin use	3	Harris, 2002 (26); Fortaleza, 2006 (25); Onguru, 2008 (27)	75–120	3	1.8–2.9
Other antibiotic use	18	2× Furtado, 2010 (55); Lepelletier, 2010 (35); 2× Martinez, 2009 (56); 2× Onguru, 2008 (27); 2× Aloush, 2006 (57); Fortaleza, 2006 (25); Zavascki, 2006 (52); Nouer, 2005 (53); Ozkurt, 2005 (41); Zavascki, 2005 (42); Defez, 2004 (54); 2× Harris, 2002 (26); Richard, 1994 (58)	15–120	9	2.2–43.7
Medical devices	21	Nagao, 2011 (59); Park, 2011 (60); Kohlenberg, 2010 (48); 2× Cezario, 2009 (36); Cortes, 2009 (61); Fortaleza 2009 (62); Martinez, 2009 (56); Pena, 2009 (49); Mueller, 2008 (37); Onguru, 2008 (27); Pereira, 2008 (63); Zavascki, 2005 (42); 2× Defez, 2004 (54); Cao, 2004 (43); 2× Dropulic, 1995 (64); Talon, 1995 (65); Thuong, 2003 (66); Lodise, Jr., 2007 (46)	6–204	13	2.1–64.3
ICU admission	8	van der Bij, 2011 (12); Lepelletier, 2010 (35); Eagye, 2009 (67); Furtado, 2009 (68); Mueller, 2008 (37); Aloush, 2006 (57); Zavascki, 2006 (52); Harris, 2002 (26)	35–120	5	1.1–13.3
Underlying disease	12	Furtado, 2010 (55); 3× Fortaleza 2009 (62); Pena, 2007 (38); 3× Zavascki, 2006 (52); Fortaleza, 2006 (25); Ohmagari, 2005 (40); Troillet, 1997 (44); Talon, 1995 (65)	17–260	6	1.0–25.0
Patient characteristics	19	Park, 2011 (60); 2× Furtado, 2010 (55); Lepelletier, 2010 (35); 2× Eagye, 2009 (67); Cezario, 2009 (36); Aloush, 2006 (57); Zavascki, 2005 (42); Ohmagari, 2005 (40); 2× Defez, 2004 (54); Berthelot, 2001 (69); Carmeli, 1999 (2); 2× Mammina, 2008 (70); 3× Montero, 2010 (47)	18–354	10	1.0–13.9
Length of hospital stay	13	Harris, 2011 (33); Furtado, 2010 (55); Lautenbach, 2010 (34); Lepelletier, 2010 (35); Yang, 2009 (50); Pereira, 2008 (63); Onguru, 2008 (27); Ozkurt, 2005 (41); Harris, 2002 (26); Carmeli, 1999 (2); 2× Montero, 2010 (47); Arruda, 1999 (71)	20–354	8	1.0–6.7
Other	18	van der Bij, 2011 (12); Harris, 2011 (33); Lautenbach, 2010 (34); Furtado, 2010 (55); Montero, 2010 (47); Pena, 2009 (49); 2× Aloush, 2006 (57); Fortaleza, 2006 (25); Zavascki, 2006 (52); 2× Ozkurt, 2005 (41); 2× Defez, 2004 (54); Paramythiotou, 2004 (72); Berthelot, 2001 (69); Carmeli, 1999 (45); Dropulic, 1995 (64)	34–354	10	1.7–13.2

^aFrom the initial 53 studies, those focused on only patients with bacteremia (n = 8) were excluded. OR, odds ratio.

^bSources are identified by first author, year, and reference number. 2× or 3×, two or three different factors per reference.

TABLE 2.

Summary of studies (n = 8) regarding P. aeruginosa bacteremia, reporting risk factors for transmission and acquisition of carbapenem-resistant P. aeruginosa, based on multivariate analyses^a

Studyc	Country	Study design	Hospital setting	No. of cases	Quality scoreb	Risk factors
						For what	Factor	OR estimate	95% CI	P value
Joo, 2011 (73)	South Korea	cc	mix	46	4	imp	Aminoglycoside use	3.60	1.39–7.31	0.025
							Urinary catheter	3.19	1.39–7.31	0.006
							Carbapenem use	2.87	1.26–6.56	0.012
							Fluoroquinolone use	2.54	1.08–5.96	0.033
Tumbarello, 2011 (74)	Italy	cc	mix	106	6	mr	Central venous catheter	17.99	6.45–50.09	<0.001
							Previous antibiotic therapy	2.79	1.10–7.07	0.03
							Corticosteroid use	2.73	1.06–7.00	0.03
Yang, 2011 (75)	South Korea	cc	pea	7	4	mr	Admission to ICU	6.82	1.3–35.8	0.023
Johnson, 2009 (76)	USA	rc	mix	113	7	mr	Hospital-acquired BSI	2.41	1.39–4.18	0.002
							Previous transplantation	2.38	1.51–3.76	<0.001
							Admission to ICU	2.04	1.15–3.63	0.015
Tam, 2007 (77)	USA	cc	mix	18	4	car	Additional wk of hospitalization	1.25	1.04–1.51	0.019
Falagas, 2006 (78)	Greece	cc	mix	16	4	mr	Carbapenem use	9.0	2.4–34.3	0.001
Kang, 2005 (79)	South Korea	rc	mix	28	6	imp	Carbapenem use	40.96	8.92–188.3	<0.001
							Fluoroquinolone use	5.60	1.64–19.11	0.006
							Invasive procedure within previous 72 h	4.51	1.56–13.04	0.005
El Amari, 2001 (80)	Switzerland	cc	mix	81	4	mr	Previous monotherapy (including imipenem)	2.5	1.3–4.8	0.006

^aOR, odds ratio; CI, confidence interval; cc case control; rc, retrospective cohort; mix, mixed; pea, pediatric general; imp, imipenem; mr, multiresistance, including carbapenems; car, carbapenem; BSI, bloodstream infection.

^bAccording to the Newcastle-Ottowa quality assessment scale.

^cStudies are reported by first author, year, and reference number.

Not all studies provided detailed information regarding the microbiological methods used. However, 23 of the 53 studies did describe the method used for the identification of P. aeruginosa, of which 10 studies used the Vitek system. Only 19 of the 53 studies described isolate genotyping, with 16 studies using pulsed-field gel electrophoresis (PFGE), 1 using multiple-locus variable-number tandem repeat analysis (MLVA), 1 using restriction fragment length polymorphism (RFLP), and 1 using repetitive-element-based PCR. The median number of cases, as included in the multivariate analyses in these 19 studies, was 30 (ranging from 6 to 204 cases). The median number of genetically identical clusters identified was 2 (ranging from 1 to 8). The median size of the clusters described in these genotyping studies was 4 (ranging from 2 to 47). Seven of the 53 studies also identified the presence of bla_VIM and bla_IMP genes (using PCR amplification). The average number of cases in these seven studies was 32 (ranging from 5 to 47 cases).

The statistically significant risk factors calculated from the multivariate analyses, specifically the presence of carbapenem-resistant P. aeruginosa (Table 1), were extracted and merged into 10 different classes. The definitions of the risk factors from the different studies were not uniform.

When considering all statistically significant risk factors from the multivariate analyses of 45 studies (n) that had not included “only bacteremic patients,” it was observed that the presence of medical devices was the most reported risk factor (n = 21) (Table 1). The risk factors extracted from the eight studies including only patients with bacteremia are shown in Table 2. Eight of the 53 studies not only identified risk factors but also identified protective factors for presence of a carbapenem-resistant P. aeruginosa, including quinolone use, exclusive feeding by formula, and duration of antibiotic treatment (Table 3).

TABLE 3.

Summary of studies reporting protective factors for transmission and acquisition of carbapenem-resistant P. aeruginosa, based on multivariate analyses

Studya	Country	Risk factor resultsb
		Risk factor	Risk estimate		95% CI	P value
van der Bij, 2011 (12)	Netherlands	Cystic fibrosis as an underlying disease	OR	0.10	0.1–0.6	NR
Fortaleza, 2009 (62)	Brazil	Quinolone use	OR	0.13	0.03–0.47	0.002
Martinez, 2009 (56)	Spain	Quinolone use	OR	0.27	0.1–0.7	NR
Martinez, 2009 (56)	Spain	Antipseudomonal cephalosporin use	OR	0.27	0.08–0.9	NR
Mammina, 2008 (70)	Italy	Exclusive feeding by formula	HR	0.18	0.05–0.61	0.006
Mammina, 2008 (70)	Italy	Length of stay of >2 weeks	HR	0.10	0.00–0.11	0.011
Lodise, Jr., 2007 (46)	USA	Risk factor 1 + 2 + 3c	PR	0.60	0.4–0.9	0.02
Aloush, 2006 (57)	Israel	Having a malignant disease	OR	0.20	0.05–0.9	0.03
Berthelot, 2001 (69)	France	Duration of antibiotic treatment	OR	0.78	0.69–0.87	NR
Arruda, 1999 (71)	Brazil	Number of antimicrobial drugs	OR	0.33	NR	0.006

^aStudies are reported by first author, year, and reference number.

^bOR, odds ratio; HR, hazard ratio; PR, prevalence ratio; CI, confidence interval; NR, not reported.

^cCombination of risk factors: 1, prior receipt of mechanical ventilation for 11 days or more; 2, prior carbapenem exposure for 3 days or more; 3, prior fluoroquinolone exposure of 3 days or more.

Possible sources and reservoirs.

Several environmental sources and reservoirs were identified (Table 4). In some outbreaks, a single source could be identified (e.g., a damaged bronchoscope or a contaminated automated urine collection machine), and the outbreak stopped after removing, repairing, or cleaning this source. However, often a reservoir was identified that was possibly not actually the main source of infection but rather a consequence of the presence of a colonized or infected patient that had led to contamination of the environment (e.g., via sinks or mattresses).

TABLE 4.

Environmental sources and reservoirs identified when searching 1,662 + 22 studies for carbapenem-resistant P. aeruginosa

Environmental source/reservoir	Reference(s)a
Automated urine analyzer	Hallin, 2012 (81); Nagao, 2011* (59)
Urine vol-measuring device	Sekiguchi, 2007 (82)
Air-conditioning system	Pinna, 2009 (83)
Sinks	Kouda, 2011 (84); Babu, 2011 (85); Crivaro, 2009 (86); Hota, 2009 (87); Mayank, 2009 (88); Crespo, 2004 (89); Boutiba-Ben Boubaker, 2003 (90); Bertrand, 2000 (91); Bert, 1998 (92); Griffith, 1989 (93)
Scopes	Boutiba-Ben Boubaker, 2003 (90). Bronchoscope, DiazGranados, 2009 (94); Sorin, 2001 (95); Panzig, 1999 (96). ERCP scope, Fraser, 2004 (97). Endoscope, Pitten, 2001 (98)
Water tap	Mentzelopoulos, 2007* (39); Bukholm, 2002 (99)
Trap water	Leung, 2008 (100)
Tap water	Mayank, 2009 (88); Pitten, 2001 (98); Bert, 1998 (92)
Sanitation related contamination	Kouda, 2011 (84); Panzig, 1999 (96); Verweij, 1997 (101)
Contaminated patient room	Kouda, 2011 (84); Cezario, 2009* (36); Mayank, 2009 (88); Boutiba-Ben Boubaker, 2003 (90); Landman, 2002 (102)
Positive cultures from nurses	Crivaro, 2009 (86); Mayank, 2009 (88); Vilar-Compte, 2003 (103); Bertrand, 2000 (91); Zheng, 1990 (104)
Bed pan sterilizer	Verweij, 1997 (101)
Milk bank pasteurizer	Gras-Le Guen, 2003 (105)
Bottle warmer	Gras-Le Guen, 2003 (105)
Stethoscope	Crespo, 2004 (89)
Mechanical ventilation related	Cezario, 2009* (36); Kikuchi, 2007 (106); Landman, 2002 (102)
Suction apparatus	Babu, 2011 (85); Mentzelopoulos, 2007* (39); Bertrand, 2000 (91)
Ice packs	Bertrand, 2000 (91)
Mops	Babu, 2011 (85)
O2 bottles, O2 tubing	Mayank, 2009 (88)
Contaminated cystoscopy room	Pena, 2003 (107)
Contaminated urodynamic lab	Climo, 1997 (108)

^aReferences are reported by first author, year, and reference number. Studies followed by an asterisk were included in the systematic review.

Study quality.

For all included studies (n = 53), a quality assessment was performed. Validation of case-control studies (n = 38) according to the Newcastle-Ottawa quality assessment scale resulted in all studies scoring between 4 and 6 stars of a possible 10 (19). However, the validation of cohort studies (n = 12) according to the Newcastle-Ottawa quality assessment scale resulted in scores between 6 and 7 stars of 13 (19). The most important reasons for not awarding a star were (i) the use of hospital controls, (ii) the use of medical records, (iii) no information about follow-up of patients, and (iv) different matching criteria between studies. Validation of the two cross-sectional studies and the single study with an observational study design, all according to STROBE guidelines, resulted in scores of 15, 17, and 18 points of a total of 22, respectively (18). The main reasons not to award points in these analyses were due to the limited description of the statistical analysis in the methods and results sections of the articles.

Nine meta-analyses.

Thirty-eight of 53 studies were included in the 9 conventional meta-analyses, reporting 106 risk factors and 5 protective factors. Eight studies were excluded because only risk factors for P. aeruginosa bacteremia were reported. Five studies were excluded because they reported hazard ratios (n = 2), risk ratios (n = 2), and a prevalence ratio (n = 1). Two studies were excluded because of missing confidence intervals. Thus, nine different meta-analyses were performed, plus an additional meta-analysis combining three risk factors (quinolone use, vancomycin use, and other antibiotic use). The results of the nine meta-analyses are shown in Table 5, and their forest plots are shown in Fig. 2. When combining the risk factors quinolone use, vancomycin use, and other antibiotic use and performing an additional meta-analysis, the pooled odds ratio was 3.07 (95% CI = 2.27 to 4.15). Publication bias indicators showed significant results for the risk factors carbapenem use, medical devices, patient characteristics, and length of hospital stay (Table 5). For the additional meta-analysis, publication bias indicators showed no significant results. Carbapenem use (OR = 7.09, 95% CI = 5.43 to 9.25) and medical devices (OR = 5.11, 95% CI = 3.55 to 7.37) resulted in the highest pooled ORs in the meta-analyses. Therefore, cumulative meta-analyses were performed for these two risk factors. Results are shown in a forest plot (Fig. 2a and b). For carbapenem use, all years showed statistically significant results. For the risk factor medical devices, the result was not significant when the estimate was updated the second time. When the estimate was updated for the third time, results became significant once more.

TABLE 5.

Conventional meta-analyses of the different risk factors for acquisition and transmission of carbapenem-resistant P. aeruginosa^a

Risk factor	No. of factors	Pooled OR (random effects)	95% CI	Range of OR in individual studies	Risk of publication bias
					Egger	P value	Kendall's tau	P value
Carbapenem use	16	7.09	5.43–9.25	3.6–76.0	1.39	0.02	0.47	0.01
Medical devices	19	5.11	3.55–7.37	2.1–64.3	2.30	<0.001	0.49	0.003
Other antibiotic use	19	3.56	2.52–5.03	0.3–43.7	1.49	0.06	0.38	0.02
ICU admission	8	3.02	1.62–5.61	1.1–13.3	2.96	0.002	0.07	0.90
Quinolone use	11	2.73	1.27–5.87	0.1–48.4	0.89	0.56	0.45	0.06
Underlying disease	13	2.44	1.23–4.84	0.1–25.0	1.34	0.06	−0.05	0.77
Vancomycin use	3	2.10	1.42–3.09	1.8–2.9	NC	NC	NC	NC
Patient characteristics	13	1.46	1.22–1.75	1.0–13.9	2.02	<0.001	0.56	0.007
Length of hospital stay	9	1.06	1.02–1.09	1.0–6.7	3.05	0.0003	0.56	0.04

^aOR, odds ratio; CI, confidence interval; NC, not calculated because there were too few strata.

FIG 2.

(a) Forest plots of conventional and cumulative meta-analyses of the risk factor carbapenem use in a random-effects model, shown on a logarithmic scale. Plots: 1, conventional meta-analysis including the source given as first author and year of publication, number of case patients (in parentheses), odds ratio, and 95% confidence interval; 2, cumulative meta-analysis including number of case patients, odds ratio, and 95% confidence interval. (b) Forest plots of conventional and cumulative meta-analyses of the risk factor medical devices using a random-effects model, shown on a logarithmic scale. Plots: 1, conventional meta-analysis including source and number of case patients as indicated for panel a, odds ratio, and 95% confidence interval; 2, cumulative meta-analysis including number of case patients, odds ratio, and 95% confidence interval. (c) Forest plots of individual and pooled odds ratios for seven different risk factors of transmission and acquisition of carbapenem-resistant P. aeruginosa, using a random-effects model, shown on a logarithmic scale.

Even when excluding cohort and cross-sectional studies (n = 8) from the meta-analyses, our estimated results changed only slightly. The mean change was +0.2, ranging from −0.1 (risk factor length of hospital stay) to +1.31 (risk factor underlying diseases). All previous significant result calculations remained significant after removal of these eight studies.

DISCUSSION

Summary of evidence.

This systematic review identified the nine most significant and most reported risk factors for the presence of carbapenem-resistant P. aeruginosa and summarized the sources and reservoirs of these bacteria within the hospital environment. The nine risk factors were, in order of statistical significance, (i) carbapenem use, (ii) medical devices, (iii) other antibiotic use, (iv) ICU admission, (v) quinolone use, (vi) underlying diseases, (vii) vancomycin use, (viii) patient characteristics, and (ix) length of hospital stay. The risk factor carbapenem use showed the strongest pooled odds ratio in the meta-analysis (Table 5). However, the most frequently reported risk factor was medical devices, which showed the second strongest pooled odds ratio (Table 5). The cumulative meta-analyses (Fig. 2a and b) of these two risk factors showed that the estimate of the risk factor carbapenem use was stable for studies published after 2005. Before 2005, only a few studies published were included, and therefore the estimate fluctuated per publication. However, after 2005, the worldwide use of carbapenem increased, mainly due to the appearance of endemic and epidemic multiresistant microorganisms, especially bacteria expressing extended-spectrum beta-lactamases in ICUs (where most of the studies included in this publication were performed) (21–24). The estimate of the risk factor medical devices decreased between 1995 and 2008 and increased from 2008 to 2011. We hypothesize that the estimate increased after 2008 due to an increase in the number of medical device days during this time period. The decrease in estimate from 1995 to 2008 can be explained by the relatively few studies included in the first part of the cumulative meta-analysis.

We also looked at whether studies identified environmental sources and/or reservoirs, not only in included studies but also in those excluded. Only 31 outbreaks reported environmental sources or reservoirs (Table 4). This implies that in most epidemics a source or reservoir is not identified, not reported, or not searched for. If carbapenem-resistant P. aeruginosa was identified in the innate environment, it was often unclear or not proven that the presumed reservoir was indeed the primary source of infection. In fact, sinks are most frequently reported and thought to be the main reservoir of carbapenem-resistant P. aeruginosa in hospitals (Table 4).

It was remarkable that in three of the studies included in the analyses, vancomycin use was identified as a risk factor for acquiring carbapenem-resistant P. aeruginosa (25–27). All three articles hypothesize that this may have been due to antibiotic selection pressure, with the reduction or elimination of competing Gram-positive bacteria post-antibiotic treatment having facilitated the colonization of the skin or gastrointestinal tract of patients with Gram-negative bacteria, including P. aeruginosa.

Limitations and strengths.

The limitations of this study are mostly related to the heterogeneity of the studies included in the analyses. From our investigations, it was obvious that every reported outbreak generally involved different target populations, microbial sources, microbiological methods, active surveillance to find cases, and methods for identifying whether there was transmission or endogenous selection.

A limitation of the meta-analyses was the diverse models used by the different studies when performing multivariate regression analysis. Also, in almost all cases, the models used were not described. This problem is already known to be a major limitation of studies utilizing meta-analyses, as “confounders” can seriously alter the combined estimate. We know that the confounders that are adjusted for are different, whereas in meta-analysis we require them to be the same. However, from a clinical point of view they have to be different, because every situation (selection or transmission), outbreak, or level of endemicity is different. Even if we knew every specific model used, it would not solve the problem of heterogeneity. For all of these reasons, we used a random-effects model.

The statistical results may also have been influenced by publication bias, and the Egger and Kendall's tau publication bias indicators showed significant results for several risk factors (Table 5). However, the authors tried to include as many studies as possible, despite differences in language or size of the outbreak. Nevertheless, a full-text article was not available for 20 studies, data were incomplete for 2 studies, and there may also be unpublished studies that we could not access. However, this number of studies is small relative to the number of studies included after title/abstract selection (n = 256), so its influence on our results is likely to be limited.

We excluded studies including only patients with CF. These patients are chronically infected with P. aeruginosa, with strains acquired mostly in the community, and are a different patient population from the population of our interest (28).

Previously, a review by Falagas and Kopterides (2006) also identified risk factors associated with P. aeruginosa infection (15). Several of the current risk factors observed (Table 1) match the risk factors observed by Falagas and Kopterides. However, in contrast to the review by Falagas and Kopterides, the current study focuses on carbapenem resistance and includes only studies that analyzed data using a multivariate model. Also, almost one-half of the studies included in this publication were published after 2006. Finally, we also included all studies that indicated a source or reservoir of their P. aeruginosa outbreak, and we conducted conventional and cumulative meta-analyses, results that are not available in the review by Falagas and Kopterides.

Conclusions and implications.

This systematic review shows that the risk factors for P. aeruginosa infection and transmission are diverse. However, the use of carbapenem antibiotics was the most significant risk estimate from this meta-analysis, which highlights the importance of antibiotic stewardship in controlling P. aeruginosa outbreaks. During an outbreak involving one or more (clonal) strains, the use of these antibiotics could be a risk factor for acquisition of that clonal strain(s), by making the patient more vulnerable to colonization or infection. Importantly, antibiotic use is a risk factor that can be influenced in order to reduce the chance of outbreaks occurring. Another risk factor is the use of medical devices and reduction of device days. The use of medical devices and the number of device days are also the most frequently reported risk factors resulting from our meta-analyses. The increased use of medical devices, and for longer periods of time, means that patients are becoming more vulnerable to acquiring MDR P. aeruginosa (29). On the other hand, other important risk factors for outbreaks involving MDR P. aeruginosa such as patient characteristics, underlying diseases, or ICU admission cannot be easily influenced.

This systematic review also shows that it is difficult to identify the actual source of P. aeruginosa outbreaks. Therefore, basic infection prevention measures remain very important. For example, contact isolation of patients and strict compliance with hand hygiene measures remain the major steps necessary to stop further transmission of outbreak isolates. This is important whether or not an assumed or proven exogenous source is responsible for the outbreak.

We believe that it is important that prospective studies relating to outbreaks of carbapenem-resistant P. aeruginosa report on sources and reservoirs of infection and that analysis of any data be performed using a multivariate statistical model. This information is extremely valuable with respect to planning future research and control measures for antibiotic-resistant P. aeruginosa. It is also very important for authors to genetically type strains associated with infection in order to identify clonal clusters of isolates. These data allow the infectious disease specialist to determine whether infection and spread are related to selection (risk factor carbapenem/antibiotic use) or transmission (e.g., risk factor medical devices). The systematic review and meta-analysis published here show the nine most important risk factors for the presence of carbapenem-resistant P. aeruginosa bacterial isolates among hospitalized patients. The identification of these risk factors is useful in controlling future outbreaks of by these organisms. In this case, risk factors such as antibiotic use and high numbers of device days have to be reduced or eliminated in order to help prevent the appearance and spread of carbapenem-resistant P. aeruginosa. In this study, carbapenem use was identified with the highest pooled odds ratio. Therefore, use of this class of antibiotics especially should be reduced.

Finally, it is important to decrease the use of antibiotics, especially the use of carbapenems, in order to help prevent resistant P. aeruginosa outbreaks. In addition, it is highly recommended that an infectious disease consultant with a broad view on the prevalence of MDR bacteria and knowledge of the most recent guidelines for antibiotic use in the hospital concerned be consulted. Indeed, studies have shown that consultation with an infectious disease consultant significantly increases the correct administration of microbiologically correct antibiotic therapy (30–32).