Predictive factors for sepsis by carbapenem resistant Gram-negative bacilli in adult critical patients in Rio de Janeiro: a case-case-control design in a prospective cohort study

Studies have investigated risk factors for infections by specific species of carbapenem-resistant Gram-negative bacilli (CR-GNB), but few considered the group of GNB species and most of them were performed in the setting of bacteremia or hospital infection. This study was implemented to identify risk factors for sepsis by CR- and carbapenem-susceptible (CS) GNB in intensive care unit (ICU) patients to improve management strategies for CR-GNB sepsis. We developed a case-case-control study from a prospective cohort of patients with systemic inflammatory response syndrome (SIRS), sepsis-2 or sepsis-3 criteria in which blood and other sample cultures were collected and antimicrobial therapy was instituted, in an adult clinical-surgical ICU, at tertiary public hospital in Rio de Janeiro, from August 2015 through March 2017. Among the total of 629 ICU admissions followed by 7797 patient-days, after applying inclusion and exclusion criteria we identified 184 patients who developed recurrent or single hospital-acquired sepsis. More than 90% of all evaluable cases of sepsis and 87% of control group fulfilled the modified sepsis-3 definition. Non-fermenting bacilli and ventilator-associated pneumonia predominated as etiology and source of CR-GNB sepsis. While Enterobacteriaceae and intra-abdominal surgical site plus urinary-tract infections prevailed in CS-GNB than CR-GNB sepsis. Carbapenemase production was estimated in 76% of CR-GNB isolates. Multivariate logistic regression analysis revealed previous infection (mostly hospital-acquired bacterial infection or sepsis) (OR = 4.28; 95% CI 1.77–10.35), mechanical ventilation (OR = 4.21; 95% CI 1.17–15.18), carbapenem use (OR = 3.42; 95% CI 1.37–8.52) and length of hospital stay (OR = 1.03; 95% CI 1.01–1.05) as independent risk factors for sepsis by CR-GNB. While ICU readmission (OR = 6.92; 95% CI 1.72–27.78) and nosocomial diarrhea (OR = 5.32; 95% CI 1.07–26.45) were factors associated with CS-GNB sepsis. The investigation of recurrent and not only bacteremic episodes of sepsis was the differential of this study. The results are in agreement with the basic information in the literature. This may help improve management strategies and future studies on sepsis by CR-GNB.

The empirical therapy of sepsis should be started within the first hour of presumed diagnosis, at a time when the clinic-epidemiological characteristics remain as the only determinants of a patient at greater risk.
Given these facts, we performed a case-case-control study to investigate predictive factors for sepsis by CR-and carbapenem-susceptible (CS) GNB in adult patients from a Brazilian public ICU. Our goal is to develop and validate a predictive score to identify patients at higher risk for CR-GNB sepsis in future studies.

Patients, setting and study design
The study followed a case-control design from a prospective cohort of patients with SIRS, sepsis-2 or sepsis-3 criteria in which blood and other samples' cultures were collected and antimicrobial therapy was instituted, for two or more days, in an adult clinical-surgical ICU, at a tertiary public hospital in Rio de Janeiro, Brazil, from August 2015 through March 2017. This study was approved by the institutional ethics committee and followed the Declaration of Helsinki and its later amendments. The study followed the STROBE recommendations for observational cohort studies (STROBE list in Additional file 1: Appendix S4) [20].
We evaluated all patients`clinical and surveillance samples cultured during the episodes, throughout the ICU stay, and the follow-up period of 30 days following the end of sepsis treatment. Surveillance cultures were not used to assess patient inclusion or exclusion in the study. Patients were not matched by any variable, considering the homogeneity of this population.
We followed all detected sepsis episodes during the ICU stay and follow-up period. Patients with more than one episode were aleatory selected in each group in such a manner that patients who presented CR-GNB sepsis were selected as case 1, patients who had CS-GNB sepsis episode were selected as case 2 and those with unknown sepsis or due to other etiologies than GNB were selected as control group. Cases and controls entered the study once and were monitored closely during the follow-up period.
We excluded patients younger than 18 years old, those with sepsis acquired in the community or associated with another healthcare institution, those who refused to sign the consent form and those suffering from polymicrobial sepsis by GNB and non GNB agents. In addition, we excluded patients initially enrolled in the control or CS-GNB case group that evolved respectively with CS-GNB or CR-GNB infection after discharge from ICU and during the followup period. The variables investigated as predictive factors were studied during the period of hospitalization prior to sepsis episode for both cases and controls. We investigated demographics and comorbidities, length of hospital stays, prior ICU hospitalization, reasons for ICU admission, simplified acute physiology score (SAPS-3) and sequential organ failure assessment (SOFA) at baseline, previous use of invasive devices and antimicrobials. The variables and their definitions are described in Tables 1 and 2. The information was collected from multiple sources including hospital records, hospital laboratory system, radiological records, hospital infection control committee daily surveillance, ICU staff daily clinical round records, and data entered daily into the ICU Epimed System. We collected and managed study data using REDCap electronic data capture tools hosted at Instituto Oswaldo Cruz, Fundação Oswaldo Cruz (IOC/ FIOCRUZ). The study formularies with the investigated variables are in Additional file 1: Appendices S1, S2 and S3. To avoid potential bias, all data collected were standardized and monitored throughout the study.  [21] sepsis-3 criteria was used retrospectively as follows: delta SOFA ≥2 between SOFA scores measured on two calendar days between the period of 72 h that preceded to 24 h that succeeded the date of initial blood culture, and on the ICU admission date; and qSOFA applied in patients without mechanical ventilation or sedation within 72 h before and 24 h after the date of blood culture.
Post-hoc analysis was performed by two research infectious disease physician investigators to review all clinical, radiological, and microbiological data. We reviewed the evidence of sepsis-2 [22] and − 3 [21], and of the infectious source. We classified the plausibility of infectious source as definitive, probable, possible or undetermined, according to Klouwenberg et al. (2013) criteria [23], adapted to include the updated Centers for Disease Control and Prevention definitions [24].

Microbiological methods
Blood cultures (aerobic and anaerobic) were processed using the BD BACTEC™ system (Becton Dickinson, Sparks, MD, EUA), according to the routine of hospital microbiology laboratory. Identification and antibiotic   [25] and European Committee on Antimicrobial Susceptibility Testing [26], including the use of meropenem, imipenem and ertapenem for all Gram-negative bacterial species, except Stenotrophomonas maltophilia which is naturally resistant to carbapenems. Carbapenemase production was investigated by phenotypic tests with phenylboronic acid and ethylene diamine tetra acetic acid. Gram-negative bacterial isolates detected in blood and other cultures were referred to Laboratório de Pesquisa em Infecção Hospitalar for microbiological confirmation by using classical and molecular biochemical methods as described in previous publications [27,28]. The search for the following genes of carbapenemases of Amber class A (bla KPC-2 ), B (bla SPM-1 , bla NDM-1, bla VIM ) and D (bla OXA-23-like , bla OXA-48-like and bla OXA-51-like ) was performed by using in-house multiplex polymerase chain reaction (PCR) test.
We classified the antimicrobial susceptibility profile of the strains in multi-drug (MDR), extensively-drug (XDR) and pan-drug resistant (PDR) and described as "possible" profiles whenever not all antimicrobials of all selective classes for each bacterial group or species were tested, according to Magiorakos et al. (2012) [29]. S. maltophilia and Burkholderia cepacia were considered MDR.

Sample size and statistical analysis
Considering an alpha error of 5%, a power of 80%, a control to case ratio of 1:1, and respectively 40 and 18% exposure to carbapenem among cases and controls [19], the sample size estimated was 152 patients.
The findings were used to build a model with clinicalepidemiological factors that can be easily identified by physicians during the first moment of patient evaluation, at a time when only clinical-epidemiological parameters can guide empirical antimicrobial therapy. All variables were analyzed using SPSS® statistics v22.0 software. Categorical variables were compared using Chi-Square or Fisher's exact test and for continuous variables, the Mann-Whitney-Wilcoxon test was used. Collinearity was investigated initially using Pearson correlation matrix and cross-tabulations between two or more variables [30]. All variables investigated as predictors were explored in univariate and multivariate logistic regression analyses using the complete data set to identify independent risk factors for CR-GNB and CS-GNB sepsis [31]. To optimize the model, we used our best knowledge not to include variables with collinearity together and give chances to those clinically meaningful. Our database has many variables that may be related to our outcome and correlated with each other. Although all variables were considered, those with small frequency and those with collinearity had to be excluded from multivariate analysis to improve the fit of the model. Possible interactions were also investigated. Using the approach described above, different models were evaluated but the best-fit model came with backward selection procedure. We also have used robust fit criteria for model comparisons (AIC and BIC) [32]. Both sides of the curve and significance level of 5% were considered in all tests.

Discussion
In this study, patients with longer length of hospital stay, previous infection, mostly hospital acquired bacterial infection, who had been previously treated with mechanical ventilation and carbapenems presented a higher risk of sepsis due to CR-GNB than control group. Whereas readmission to the ICU and prior nosocomial diarrhea were factors associated with CS-GNB sepsis in the study population.
Previous infection has been rarely reported as risk for infection/colonization or bacteremia by CR A. baumannii [7,8]. In our cohort, previous infection was the most important factor for CR-GNB sepsis, mainly of bacterial origin, mostly nosocomial infection and 48% previous sepsis. Recurrent pattern of sepsis was a striking feature of our studied population. Therefore, more attention for the prevention and control of nosocomial sepsis is required for the prevention of subsequent hospital-acquired CR-GNB sepsis, especially caused by A. baumannii, at the same hospitalization period in ICU patients.
The studies that have evaluated risk factors for CR-GNB or CR A. baumannii bacteremia, SIRS or nosocomial infection have found mechanical ventilation, respiratory failure, but also VAP as risk factors in critically ill patients [7,8,18,19]. In our study we did not investigate types of infection as predictive factors because we sought to investigate variables that would readily discriminate patients with increased risk, in order to be useful to guide empirical therapy in future.
Prolonged hospital stay is a classically recognized risk for hospital and ICU infection [10,11,14,19,33]. We found a 2% increase in the chances of developing sepsis by CR-GNB for every day of hospitalization. ICU readmission is also a documented risk factor for the acquisition of CS-GNB infection [10,16]. A few studies have found diarrhea to be associated with GNB bacteremia [34][35][36].
As far as we know, this is the first risk factor study in CR-GNB sepsis that has considered repetitive episodes and not only bacteremia, but the broad variety of infection sources commonly observed in ICU sepsis. Studies that have investigated risk factors for infections by CR-GNB species are still rare [18,19]. In general, the studies focus on risk factors for infections by specific species such as K. pneumoniae [12,13,16] or other members of the Enterobacteriaceae [15,17], P. aeruginosa [10,11] and A. baumannii [7][8][9], and most of them in hospital infection [9,10,12,[15][16][17] or bacteremia [8,11,18,19]. However, bacteremia has been detected in less than 30% of septic cases in ICU [37,38]. The extensive use of vascular catheter has been recognized as the most important factor contributing to BSI [4,39,40], while VAP has predominated as source of sepsis in ICU [38]. Therefore, although bacteremia can be considered the gold standard it represents only part of the population who were diagnosed with sepsis in ICU and is frequently associated with vascular catheter infection. In addition, several risk factor studies select the first episode of infection or bacteremia only [8-10, 13, 16-19]. The successful longitudinal follow-up of this cohort study allowed identifying patients who presented recurrent sepsis, whose diagnoses were also essential to better select cases and controls. Giving the chance of inclusion of all episodes and the variety of sepsis infections may be more reliable to better discriminate patients at higher risk in ICU.
The case-case-control design is more effective for the identification of risk factors for antimicrobial resistant pathogens, avoiding bias of exposure to the antimicrobial of interest, for not using CS-GNB sepsis as control group [41,42]. Our study aimed to confirm exposure to carbapenems as a risk for the development of CR-GNB sepsis [11,12,18,19,43], since our control group was composed of patients with sepsis and antimicrobial treatment. This methodology allows the application of the factors found in the management of empirical therapy, which would not be possible if the control was a patient without sepsis and antimicrobial treatment of the episode investigated [42].
The design resulted in the selection of cases so efficiently that the study detected a remarkable difference in the etiologies of sepsis by CR-GNB and CS-GNB. Indeed, the CR-GNB sepsis presented a predominance of non-fermenting bacteria, mainly A. baumannii, which was likely related to the higher prevalence of respiratory tract infections as source of sepsis [18,19]. Whereas Enterobacteriaceae species predominated among CS-GNB sepsis, mostly associated with respiratory tract infections, but also intra-abdominal SSI and UTI.
Evaluated sepsis episodes in cases and control patients had a good agreement when considering the adapted parameters for the diagnosis of sepsis-3 [21]. The plausibility of infectious source as definitive and probable occurred in a high proportion in all groups of cases while the source of infection was determined in the majority of controls.
Our data characterize sepsis by CS-GNB mostly as hospital-acquired infection outside the ICU, whereas sepsis by CR-GNB was mainly ICU-acquired infection. This issue corroborates with a well-known fact that ICU patients have more risk of infection by resistant bacteria [33]. In fact, CR-GNB species typify a bacterial population with a high level of resistance and few treatment options, due to the often high carbapenems MIC and some combined resistance to polymyxins. Molecular investigation of carbapenemase production shows that we have determined the risk factors for CR-GNB sepsis predominantly with this mechanism of resistance, which is commonly described and disseminated worldwide [43][44][45][46][47].
Main limitation of this study may be described as a single center study, which indicates caution to any generalization of our findings. Adapting management strategies to the local epidemiological data is a general recommendation for the prevention and control of hospital-acquired infection, which indicates the performance of cyclic evaluations locally. The methodology of case-control selection was one of the study's strengths. The performance of microbiological methods and the use of antimicrobials, which may have inhibited microbial growth in clinical cultures, could have influenced the risk factors results. Other non-investigated elements may also have interfered. Phylogenetic analysis of GNB isolates would have improved our knowledge about the epidemiological context of GNB hospital-acquired sepsis.

Conclusions
Prolonged hospitalization with the development of healthcare-associated infection, requiring mechanical ventilation and treatment with carbapenems seem to be the natural history for subsequent sepsis by carbapenemase-producing GNB in this population. The concordance with the background knowledge suggests that these factors should be evaluated further for developing and validating a risk score to identify patients at higher risk for CR-GNB sepsis in ICU. Little is known about the influence of recurrent sepsis during the same hospitalization period. Consequently, investigation of repetitive episodes of sepsis in ICU patients is warranted.