Characterization and Clinical Impact of Bloodstream Infection Caused by Carbapenemase-Producing Enterobacteriaceae in Seven Latin American Countries

Introduction Infections caused by carbapenem-resistant Enterobacteriaceae are a public health problem associated with higher mortality rates, longer hospitalization and increased healthcare costs. We carried out a study to describe the characteristics of patients with carbapenemase-producing Enterobacteriaceae (CPE) and non-CPE bloodstream infection (BSI) from Latin American hospitals and to determine the clinical impact in terms of mortality and antibiotic therapy. Methods Between July 2013 and November 2014, we conducted a multicenter observational study in 11 hospitals from 7 Latin American countries (Argentina, Colombia, Ecuador, Guatemala, Mexico, Peru, Venezuela). Patients with BSI caused by Enterobacteriaceae were included and classified either as CPE or non-CPE based on detection of blaKPC, blaVIM, blaIMP, blaNDM and blaOXA-48 by polymerase chain reaction. Enrolled subjects were followed until discharge or death. Demographic, microbiological and clinical characteristics were collected from medical records. Both descriptive and inferential statistics were used to analyze the information. Results A total of 255 patients with Enterobacteriaceae BSI were included; CPE were identified in 53 of them. In vitro non-susceptibility to all screened antibiotics was higher in the patients with CPE BSI, remaining colistin, tigecycline and amikacin as the most active drugs. Combination therapy was significantly more frequent in the CPE BSI group (p < 0.001). The most common regimen was carbapenem + colistin or polymyxin B. The overall mortality was 37% (94/255). Overall and attributable mortality were significantly higher in patients with CPE BSI (p < 0.001); however, we found that patients with CPE BSI who received combination therapy and those who received monotherapy had similar mortality. After multivariate adjustment, CPE BSI (adjusted odds ratio [aOR] 4; 95% confidence interval [CI] 1.7–9.5; p = 0.002) and critical illness (aOR 6.5; 95% CI 3.1–13.7; p < 0.001) were independently associated with in-hospital mortality. Conclusions This study provides valuable data on the clinical characteristics and mortality risk factors in patients with CPE BSI. We determined that CPE infection is an independent mortality predictor and thus Latin American hospitals should perform campaigns on prevention and control of CPE BSI.


Introduction
Infections caused by carbapenem-resistant Enterobacteriaceae (CRE) have been increasingly reported in the last decade and have become a public health problem of global dimensions [1][2][3]. Carbapenemases constitute the major mechanism of carbapenems and most other β-lactam antibiotics resistance in Enterobacteriaceae [1,4,5]. CRE are also commonly associated with additional mechanisms of resistance to other antibiotic classes, including aminoglycosides and fluoroquinolones [6,7]. As a result of this broad-spectrum antibiotic resistance, very limited options for treating CRE infections are available [1].
While previous Latin American studies have focused on the molecular characterization of carbapenemases in Enterobacteriaceae and its epidemiological surveillance [5,22,23], few regional studies on the clinical impact of carbapenemase-producing Enterobacteriaceae (CPE) BSI have been performed [18]. In this study, we aimed to describe the clinical characteristics of patients with CPE and non-CPE BSI from Latin American hospitals, and to determine the clinical impact in terms of in-hospital mortality and antibiotic therapy.

Methods Setting
The study was conducted between July 2013 and November 2014 in 11 high-complexity medical centers from Argentina, Colombia, Ecuador, Guatemala, Mexico, Peru, and Venezuela.

Study design and population
This was a multicenter observational study that included consecutive inpatients of any age and sex with BSI caused by any Enterobacteriaceae with minimal inhibitory concentration (MIC) 1 μg/mL for cefotaxime and/or ceftriaxone. Only one isolate per patient was included. Individuals included had clinical findings of systemic inflammatory response syndrome [24]. Exclusion criteria were patients whose isolates grew more than one microorganism or whose patient's clinical data were incomplete. Patients with CPE BSI were designated as exposed patients, whereas patients with non-CPE BSI were designated as non-exposed. CPE was defined based on the molecular detection by polymerase chain reaction (PCR) of any carbapenemase (described ahead).

Sample size and sampling
Sample size was calculated using Stata 1 version 9.0 (StataCorp LP, College Station, TX, USA). We considered a mortality rate of 32.1% in patients with CPE infection and of 9.9% in patients with non-CPE infection [25], a bla KPC prevalence of 12.8% in gram-negative bacteria [26], a 95% confidence level and an 80% power. Ratio of exposed to non-exposed patients was 1:4.

Data collection
Enrolled subjects who met selection criteria were followed until discharge or death. Demographic, microbiological and clinical characteristics (i.e., age, sex, acquisition of infection, bacterial isolate, bacteremia source, Pitt bacteremia score (PBS) [27], intensive care unit [ICU] admission, underlying diseases and comorbidities, and treatment of the bacteremic episode) were collected from medical records into a case report form (CRF). Blood samples obtained during hospitalization were used for microbiological cultures. All microbiology laboratories of the participating hospitals used automated systems for microorganism identification and in vitro antibiotic susceptibility testing. Antimicrobial MICs were interpreted according to the breakpoints of the Clinical and Laboratory Standards Institute (CLSI) 2014 guidelines [28]. In each hospital one co-investigator and one coordinator were in charge of the patient selection and follow-up, as well as of the data collection, and shipment of isolates and CRFs to the International Center for Medical Research and Training (CIDEIM).
Treatment regimens were classified either as monotherapy (one in vitro active antibiotic) or combination therapy (two or more in vitro active antibiotics). We determined the final outcome as discharge or all-cause death at 72 hours, 7 days and 28 days since bacteremia onset. Patients discharged before the respective cutoff were considered survivors. Mortality attributable to BSI was defined as the death of a patient with clinical and laboratory evidence of ongoing infection in absence of other feasible reasons.

Statistical analysis
We conducted database processing and analysis in Stata 1 version 9.0 (idem) (Dataset in S1 Dataset). Prior to analysis, data was anonymized, de-identified and aggregated. We used relative frequencies for categorical variables, and central tendency and dispersion measures for numerical variables. Data were analyzed between exposed and non-exposed subjects using Chi-square (χ 2 ) test or Fisher exact test for categorical variables, and Student t test or Mann-Whitney U test for numerical variables, as appropriate. All p values were deemed as statistically significant when < 0.05. Time-to-event analysis was performed through Kaplan-Meier estimates between groups, and these were compared by the log rank test at 72 hours, 7 days and 28 days after bacteremia onset. Moreover, variables included in the multivariate logistic regression model were those with p value < 0.2 in bivariate analysis (critical illness, change of the empirical regimen after culture report, definitive treatment with carbapenem, CPE BSI). Independent predictors of in-hospital mortality were sought. Odds ratios (ORs) and their 95% CIs were calculated.

Ethical approval
The study was performed in accordance with the 1964 Declaration of Helsinki and its later amendments.

General characteristics of patients and isolates
Six hundred and eighty-six Enterobacteriaceae blood isolates were collected and sent to CIDEIM, among which 255/686 (37%) with their corresponding patients' CRFs met selection criteria and hence were included in the study. Distribution per country of Enterobacteriaceae isolates and CRFs included is shown in Table 1. CPE were identified in 53/255 (21%) of all individuals; the other 202 patients had non-CPE isolates ( Table 2). Out of the 53 CPE isolates, 44 (83%) harbored bla KPC , 5 (9%) harbored bla VIM , and 4 (8%) harbored bla NDM . None of the isolates were positive for IMP or OXA-48. Interestingly, all NDM isolates were from Guatemala and all VIM isolates were from Mexico.

Antibiotic treatment
Out of the 231 patients in whom empirical antibiotic treatment was recorded, 114 (49%) received at least 1 active antibiotic, while 117 (51%) received no active antibiotic. Definitive treatment was given to 228/255 (89%) patients; 212 (93%) of these 228 patients received at  Table 2. The proportion of empirical and definitive active antibiotic treatment was significantly higher in the patients with non-CPE BSI (p < 0.001). Combination therapy was significantly more frequent in the CPE BSI group than in the non-CPE BSI group (p < 0.001). There were no statistically significant differences in the change of the empirical antibiotic regimen after culture report and the use of a carbapenem in the definitive regimen among study groups. The patients with CPE BSI who received definitive antibiotic treatment were 48/53 (91%) ( Table 4) but only 37 of these 48 received an active antibiotic regimen. While 29/37 (78%) received a combination therapy, 8/37 (22%) received monotherapy. A carbapenem was part of the regimen in 35 (95%) of those 37 patients. Among the patients who received active combination therapy, the most common regimen was carbapenem + colistin or polymyxin B (11/29, 38%).

Mortality and survival
The overall mortality was 37% (94/255). Mortality was significantly higher in patients with CPE BSI than in those with non-CPE BSI (64% vs. 30%, p < 0.001). Mortality was attributable to BSI in 55/94 dead individuals (59%). Attributable mortality was significantly higher in the patients with CPE BSI (85% vs. 43%, p < 0.001). In the CPE BSI group, the mortality of patients who received an active antibiotic regimen did not statistically differ from the mortality of the patients without an active antibiotic therapy (22/37 vs. 8/11, p = 0.5). Mortality in monotherapy group was similar to mortality in the combination therapy group (5/8 vs. 17/29, p = 1.0). No significant difference was found between subjects with CPE BSI and subjects with non-CPE BSI regarding Kaplan-Meier survival analysis at 72 hours (p = 0.07). In contrast, there was a significantly greater survival in non-critical patients than in critical patients at 72 hours (p < 0.0001). Kaplan-Meier survival estimates at 7 days were significantly better for patients with non-CPE BSI than for those with CPE BSI (p < 0.001) (Fig 1). Also, there was a statistically significant difference in survival of the patients at day 7 according to critical or non-critical illness (p < 0.0001). Similarly, significant survival differences were found at 28 days for these outcomes (p < 0.0001). In multivariate analysis (Table 5), we found that CPE BSI (adjusted OR [aOR] 4; 95% CI 1.7-9.5; p = 0.002) and critical illness (aOR 6.5; 95% CI 3.1-13.7; p < 0.001) were independently associated with in-hospital mortality.

Discussion
Since 1990s, CRE have been reported worldwide [1,32,33]. CRE poses a public health threat given their alarming endemicity in many world regions, including Latin America [5], and its association with worse clinical and economic outcomes [1,2,8]. In this Latin American study of patients with BSI caused by Enterobacteriaceae, we described the clinical characteristics and determined the risk factors associated with in-hospital mortality. Of the 255 Enterobacteriaceae isolates included, 53 were found by PCR to harbor KPC, VIM or NDM. The molecular characterization performed to the isolates was essential because the phenotype reported by automated systems routinely used in microbiology laboratories might not detect up to 6-87% of the  β-lactamases associated with carbapenem resistance in known resistant isolates [34]. As many CPE just have discretely elevated MICs [35], PCR was a reliable technique for the classification of the patient groups in this study. KPC is the most prevalent Ambler class A carbapenemase worldwide and has the biggest clinical significance [5,10]. In this study, most CPE (44/53, 83%) were KPC producers, and the majority of the CPE isolates (39/53, 73%) corresponded to K. pneumoniae, which is the most frequently identified carbapenemase producer. This finding is consistent with other studies [1,5].
As expected, CPE isolates presented a higher in vitro resistant profile to all antibiotics. Among the carbapenems, resistance to ertapenem was present in 95% of the strains, while resistance rate to imipenem and meropenem was 74% and 78%, respectively. CPE isolates tested for doripenem exhibited 100% resistance although the interpretation was limited given the small number of isolates tested (n = 9). Colistin, tigecycline and amikacin had the best in vitro activity against CPE, with a susceptibility proportion of 96%, 79% and 78%, respectively.
Both the proportions of empirical and definitive active antibiotic treatment were significantly lower in patients with CPE BSI, likely due to the few antibiotic options for which CPE are susceptible. For the same reason, combination therapy was likely found to be more frequent in patients with CPE BSI. Among 48 patients who received definitive antibiotic treatment, 29 received an active combination therapy. Carbapenems were used in 28 (97%) of these antibiotic regimens.
Mortality in patients with CPE BSI was 64% which is similar to other studies carried out in the USA and Europe [8,[11][12][13][14][15][16][17]20,21]. Only one study addressing CPE infections has been performed in Latin America, in which 118 patients infected with KPC-producing Enterobacteriaceae were included; 78 patients of these presented bacteremia and the 30-day mortality was 50% [18]. In our study, in which all patients had bacteremia not exclusively caused by CPE, we found that both overall and attributable mortality rates were higher in patients with CPE (p < 0.001).
Patients who received monotherapy and patients who received combination therapy had similar mortality rates (63% vs. 59%; p = 1.0). This finding is interesting as other research works on patients infected with CRE elsewhere have demonstrated lower mortality when treated with combination therapy instead of monotherapy [15,16,38]. In our study, we probably found a non-statistically significant association due to the low number of patients with CPE BSI (n = 53) and also due to the study design. However, this is a subject of wide debate and stands yet as an unresolved question.
In our study, the survival analyses clearly showed that the adverse outcome likely occurred more frequently at 7 days after bacteremia onset among patients with CPE BSI and among critically ill patients. Although Kaplan-Meier estimates for survival were also significantly higher in patients with non-CPE BSI even at 28 days, analysis at that time is not the most appropriate because several causes not evaluated in our study could lead to in-hospital death.
Although we were mainly interested in determining if CPE BSI was a risk factor for mortality, we first assessed the presence of underlying diseases and comorbidities in the patients with CPE BSI and patients with non-CPE BSI. Most underlying diseases were comparable between both groups; however, critical illness (PBS 4) was statistically significant higher in the CPE BSI group. We found that patients presented significantly more hypotension and mechanical ventilation although the other PBS parameters were comparable among patient groups. After multivariate adjustment, CPE BSI and critical illness were statistically significant factors associated with in-hospital mortality. This indicates that among patients with BSI, those caused by CPE and those with critical illness are independently at a higher risk of death during hospitalization.
There are some limitations in this study to discuss. First, 431/686 (63%) of the isolates received were excluded because they did not meet the selection criteria; this could cause some undetermined bias in our results. Second, different hospitals had no susceptibility testing of the isolates for colistin and polymyxin B, limiting our conclusions on their effectiveness, considering that these antibiotics are often the last therapeutic option for CRE infections. Third, despite that empirical and definitive treatments were assessed in our study, time to the antibiotic infusion and removal of the infection source (i.e., catheter removal, drainage, and debridement) were not evaluated. Patel et al showed higher survival in patients infected with carbapenemresistant K. pneumoniae treated with antibiotics and removal of the focus, compared with those treated with antibiotics alone, although they did not adjust for illness severity [2].
In conclusion, this study provides valuable regional data on the characteristics and clinical impact on mortality of patients with CPE BSI from high-complexity hospitals of Latin America. We confirm that CPE infection is an independent mortality predictor which could encourage healthcare settings to carry out campaigns on prevention, detection, control and treatment of CPE BSI. Finally, we highlight the need to improve actions towards antimicrobial stewardship practices, and increase awareness on CPE among all the public health decision-makers.