Clinical Infectious Diseasescid.oxfordjournals.org
Factors Associated with Candidemia Caused by Non-albicans Candida Species Versus Candida albicans in the Intensive Care Unit
- Jennifer K. Chow1,*,
- Yoav Golan1,
- Robin Ruthazer2,
- Adolf W. Karchmer3,
- Yehuda Carmeli3,
- Deborah Lichtenberg1,
- Varun Chawla3,
- Janet Young1, and
- Susan Hadley1
+ Author Affiliations
- Reprints or correspondence: Dr. Jennifer Chow, Div. of Geographic Medicine and Infectious Diseases, Tufts—New England Medical Center, 750 Washington St., Box 041, Boston, MA 02111 (jchow@tufts-nemc.org).
Abstract
Background. Candida albicans has been the most common cause of fungal bloodstream infections (BSIs) in intensive care units (ICUs); however, infections due to non-albicans Candida species have been increasing in prevalence. We examined factors associated with BSIs due to non-albicans Candida species, compared with C. albicans BSIs, in an ICU patient population.
Methods. For our case-comparator study, we identified consecutive adult ICU patients with BSIs due to non-albicans Candida species or C. albicans at 2 tertiary care hospitals during the period 1995–2005. Data collected included demographic characteristics, comorbidities, exposure to antibiotics and antifungals, and ICU-related factors, such as total parenteral nutrition, blood product transfusions, invasive procedures, central venous catheter use, hemodialysis, and mechanical ventilation. We built a multivariable logistic regression model that identified variables that differentiate BSIs due to non-albicans Candida species from BSIs due to C. albicans.
Results. There were 67 patients with BSIs due to non-albicans Candida species and 79 patients with C. albicans BSIs. Variables were adjusted for time at risk. In multivariable models, factors associated with an increased risk of BSIs due to non-albicans Candida species, compared with C. albicans BSIs, included fluconazole exposure (odds ratio, 11.6; 95% confidence interval, 2.28–58.8), central venous catheter exposure (odds ratio, 1.95; 95% confidence interval, 1.10–3.47), and mean number of antibiotics per day (odds ratio, 2.31; 95% confidence interval, 0.71–7.54). Total parenteral nutrition exposure was associated with a decreased risk (odds ratio, 0.16; 95% confidence interval, 0.05–0.47) of BSIs due to non-albicans Candida species, compared with C. albicans BSIs. Duration of stay in the ICU was not significantly different between the 2 groups. Specific antibiotics, such as vancomycin and piperacillin-tazobactam, were not independently associated with BSI due to non-albicans Candida species.
Conclusions. Receipt of fluconazole and central venous catheter exposure were associated with an increased risk of BSI due to non-albicans Candida species, and total parenteral nutrition was associated with a decreased risk of BSI due to non-albicans Candida species, compared with BSI due to C. albicans. Patients without characteristics of infection due to non-albicans Candida species might benefit from empirical antifungal therapy with fluconazole.
Candida species are the fourth leading cause of nosocomial bloodstream infections (BSIs) in the United States [1] and are the third most common cause of BSIs in intensive care units (ICUs), accounting for 10% of such infections [2]. Currently, more than one-half of candidemia cases occur in medical or surgical ICUs [3]. Admission to an ICU is a risk factor for candidemia [4, 5], with mortality rates approaching 50% in this patient population [2, 6]. Moreover, infection due to Candida species is associated with increased durations of ICU and hospital stays, resulting in higher health care costs [7].
Although Candida albicans continues to be the most common cause of candidal BSIs, the epidemiology of species causing candidemia is changing. Recent longitudinal studies have observed an increasing proportion of BSIs caused by non-albicans Candida species, such as Candida glabrata, Candida krusei, Candida parapsilosis, and Candida tropicalis, which currently account for approximately one-half of all cases of candidemia [1, 8–14]. There are, however, limited data regarding factors associated with such infections [8, 11]. Decreased in vitro susceptibility to fluconazole is more common among several non-albicans Candida species; however, the clinical relevance of this fact is not well defined [15, 16]. However, C. albicans remains the most common species and is generally susceptible to fluconazole. With the availability of echinocandins—which are more active against several non-albicans Candida species but also are more expensive—clinicians need to decide when fluconazole is still safe for the empirical treatment of candidemia. Information that helps differentiate patients with C. albicans candidemia is of value, because these patients can be safely treated with fluconazole. One approach to this question has been through rapid laboratory identification of C. albicans, compared with non-albicans Candida species, directly from blood culture bottles with use of the peptide nucleic acid–fluorescent in situ hybridization test [17]. This test has been shown to be cost-saving for hospitals by avoiding unnecessary use of caspofungin; however, this assay is not widely available and is costly [18]. An alternative approach to providing guidance in the decision of empirical anticandidal therapy is to examine risk factors that distinguish C. albicans infections from infections due to non-albicans Candida species using clinically available data, which was the aim of our study.
When a preliminary blood culture shows growth of a Candida species, clinicians require additional guidance in deciding whether the patient can be treated safely with fluconazole. Understanding the characteristics that differentiate critically ill patients with BSIs due to non-albicans Candida species from such patients with C. albicans BSIs will assist clinicians with this early treatment decision for candidemia. Moreover, defining patients at increased risk of developing invasive infections due to different Candida species will provide important background information for the design of future clinical trials of antifungal empirical treatment. In this study, we report factors that differentiate BSIs due to non-albicans Candida species from BSIs due to C . albicans in an ICU patient population.
Patients and Methods
Study design and study population. A case-comparator study of medical and surgical ICU patients at 2 urban, university-based hospitals was conducted. Patients with Candida BSIs were identified from the microbiology laboratory log at Tufts–New England Medical Center (Boston, MA; 420 beds) during the period 1995–2005 and from the electronic microbiology laboratory database at Beth Israel Deaconess Medical Center (Boston, MA; 550 beds) during the period 1998–2004. To eliminate selection bias, we included all consecutive ICU candidemia cases at both study sites. For patients with multiple candidemia episodes, only the first episode was included. To only study patients who acquired candidemia during their stay in the ICU, we excluded cases that developed within the first 48 h after ICU admission. Patients infected simultaneously with both C. albicans and non-albicans Candida species were also excluded. Because our primary aim was to identify factors associated with having BSI due to non-albicans Candida species to help guide early empirical antifungal therapy for patients who have already acquired candidemia, we used patients with BSIs due to non-C. albicans Candida species and patients with C. albicans BSIs as the 2 comparison groups. Each center's institutional review board approved data collection procedures.
Data collection. Trained study team members collected demographic and clinical data by chart review. For the ICU stay, we collected data that would be readily available to clinicians caring for ICU patients. Data collected included patient demographic characteristics; pre-ICU operations; comorbidities; concomitant infections; exposure to antibiotic, antifungal, and immunosuppressive therapy; total parenteral nutrition (TPN); blood product transfusions; gastrointestinal procedures and operations; central venous catheter (CVC) use; hemodialysis; mechanical ventilation; and in-hospital mortality. Gastrointestinal procedures included endoscopic retrograde cholangiopancreatography, biliary tubes, percutaneous endoscopic gastrostomy tubes, and abdominal surgical procedures. Enteric bacteremia included BSIs due to Enterococcus species, gram-negative bacilli, or Bacteroides species. Immunosuppressants included both corticosteroids and transplant-related medications. Anti-anaerobic antibiotics included clindamycin, metronidazole, cefotetan, cefoxitin, carbapenems, and penicillin–²-lactamase combinations.
Risk factors were assessed from the day of ICU admission until the day before the positive blood sample was obtained (i.e., time at risk). To better quantify continuous exposures, such as receipt of fluconazole therapy, and to adjust for time at risk (rather than using a crude binary variable of yes or no), we used the proportion of time at risk during which the patient was exposed to the risk factor. The purpose for the use of antimicrobial agents, including fluconazole, was not discerned.
Subgroup analyses. Our primary aim was to compare patients with candidemia due to non-albicans Candida species with patients with C. albicans candidemia. Because non-albicans Candida species can have heterogeneous patterns of azole susceptibility, we grouped non-albicans Candida species more likely to have decreased azole susceptibility (i.e., C. glabrata and C. krusei) separate from other non-albicans Candida species (i.e., C. tropicalis, C. parapsilosis, Candida lusitaniae, and Candida dubliniensis) [15, 16]. We compared patients with C. glabrata or C. krusei BSIs with patients with C. albicans BSIs, and we compared patients with C. tropicalis, C. parapsilosis, C. lusitaniae, or C. dubliniensis BSIs with patients with C. albicans BSIs.
Statistical analysis. Factors associated with candidemia due to non-albicans Candida species, compared with C. albicans candidemia, were examined using χ2 tests or univariate logistic regression. Variables that had statistical significance at P<.30 in the univariate analysis were considered to be candidates for the building of multivariable models.
Stepwise and best subsets approaches were used to build multivariate logistic regression models to determine which variables were most strongly associated with BSIs due to non-albicans Candida species. The models contained variables that were a priori considered to be clinically relevant to avoid finding spurious associations. Because risk factors were adjusted for time at risk, which is closely correlated with duration of stay in the ICU, we did not include duration of stay in the ICU in the final model. A P value of <.05 was considered to be statistically significant in the multivariable modeling.
We assessed model discrimination by constructing a receiver operating characteristic curve, and the area under the receiver operating characteristic curve was calculated using the c statistic. Models were compared using the c statistic and χ2 tests comparing model likelihood ratios. All statistical analyses were performed in SAS, version 9.1 (SAS Institute).
Results
Patient characteristics. A total of 146 patients were included, 67 of whom had fungemia due to non-albicans Candida species and 79 of whom had fungemia due to C. albicans. The non-albicans Candida species isolated included 33 C. glabrata (49%), 13 C. tropicalis (19%), 12 C. parapsilosis (18%), 4 C. lusitaniae (6%), 3 C. krusei (5%), and 2 C. dubliniensis (3%). The median times from admission to the ICU until the onset of non-albicans Candida BSI and C. albicans BSI were 10 and 11 days, respectively. Patients with BSIs due to non-albicans Candida species tended to be older (P=.09) and had longer durations of hospitalization (P=.06), compared with patients with C. albicans BSIs; however, these trends did not reach statistical significance. No statistically significant differences with regard to sex, admission service, time from hospital admission to ICU transfer, duration of stay in the ICU, or comorbidities were found. Many patients were immunocompromised by diabetes, cirrhosis, chronic renal disease, HIV infection, solid or hematologic malignancy, or hematopoietic stem cell or solid organ transplantation; however, there were no differences by immunosuppressive disease etiology between patients with BSIs due to non-albicans Candida species and those with C. albicans BSIs (table 1).
Table 1
Demographic characteristics and comorbidities of intensive care unit (ICU) patients with bloodstream infections (BSIs) due to non-Candida albicans Candida species or C. albicans.
Treatment with antibiotics, the presence of CVCs, mechanical ventilation, and RBC transfusions were noted frequently in both groups. More patients with BSIs due to non-albicans Candida species were exposed to antifungal agents. Among patients who received antifungal therapy prior to developing candidemia, the majority received fluconazole and a few patients received amphotericin (of any formulation), caspofungin, or voriconazole. In-hospital mortality approached 60% in both groups (table 2).
Table 2
Intensive care unit (ICU)—related characteristics of patients with bloodstream infections (BSIs) due to non-albicans Candida species or Candida albicans.
Factors associated with candidemia due to non–C. albicans Candida species. After univariate analysis, the following 9 variables were considered to be candidates for the multivariate model: age, preexisting lung disease, gastrointestinal procedures, number of antibiotics administered, number of platelet transfusions, proportion of days of receipt of fluconazole therapy, and duration of CVC use, TPN, and corticosteroid therapy (table 3). The number of antibiotics per day, but not duration of antibiotic exposure (OR, 0.99; 95% CI, 0.73–1.35), was associated with an increased risk of developing BSIs due to non-albicans Candida species in univariate analysis. The type of antibiotic, such as vancomycin, piperacillin-tazobactam, or anti-anaerobic agents, was not significantly associated with the development of candidemia due to non-albicans Candida species. Concomitant bacteremia, use of immunosuppressive therapy, mechanical ventilation, hemodialysis, and RBC transfusions were not associated with candidemia due to non-albicans Candida species.
Table 3
Description of adjusted variables used for logistic regression models.
Multivariate logistic regression analysis. In the final multivariate model, proportion of days of receipt of fluconazole therapy and CVC duration were associated with an increased risk of BSI due to non-albicans Candida species, compared with C. albicans BSI. The number of antibiotics per day was also associated with an increased risk of BSI due to non-albicans Candida species, compared with BSI due to C. albicans, but the difference did not reach statistical significance. Duration of TPN, in contrast, was associated with a decreased risk of BSI due to non-albicans Candida species, compared with C. albicans BSI (table 4). No independent associations were found with age, lung disease, gastrointestinal procedures, duration of corticosteroid therapy, or number of platelet transfusions.
Table 4
Results of univariate and multivariate logistic regression analyses of factors associated with bloodstream infections (BSIs) due to non—Candida albicans Candida species, compared with those associated with BSIs due to C. albicans, in 146 intensive care unit patients.
Calibration and discrimination of the model were adequate. The area under the receiver operating characteristic curve for the model was 0.73. There were no statistically significant, clinically meaningful interactions among variables in the multivariate model. To ensure that the final model was not driven by a center effect, a categorical variable for hospital was added to the final model. This hospital variable did not significantly change the overall model and was not statistically significant (data not shown). To examine the temporal distribution of both BSIs due to non-albicans Candida species and C. albicans BSIs, the cases of candidemia were separated by hospital and were graphed by date of infection. These graphs demonstrate that candidemia occurred randomly throughout time and that there was no nosocomial clustering (data not shown).
Additional analyses. The relationship between fluconazole exposure and increased risk of candidemia due to non-albicans Candida species was dose dependent in a linear fashion. Neither a threshold nor a plateau effect was found. In other words, a single dose of fluconazole conferred increased risk of candidemia due to non-albicans Candida species, and this risk increased with increasing proportion of days of fluconazole exposure. Subgroup analyses of patients with C. albicans infection (n=79), compared with patients with C. glabrata and C. krusei infections (n=36), and of patients with C. albicans infections, compared with patients with infections due to non-albicans Candida species other than C. glabrata and C. krusei (n=18; i.e., C. tropicalis, C. parapsilosis, C. lusitaniae, and C. dubliniensis), demonstrated results similar to those of the comparison of patients with C. albicans infection with patients infected with all non-albicans Candida species grouped together, particularly for proportion of days of fluconazole exposure (OR for C. glabrata or C. krusei vs. C. albicans, 9.41; P=.02; OR for C. tropicalis, C. parapsilosis, C. lusitaniae, or C. dubliniensis vs. C. albicans, 11.6; P=.01).
Discussion
Although C. albicans remains the predominant etiology of fungal BSIs, recent epidemiologic studies of candidemia have demonstrated an increasing incidence of infections due to non-albicans Candida species in the United States and Europe [1, 8–14, 19]. Prior studies of risk factors for candidemia due to non-albicans Candida species have examined limited populations, such as those with hematologic malignancies [20, 21], or have focused on single species, such as C. glabrata, C. parapsilosis, or C. krusei, and have not looked at all non-albicans Candida species grouped together [5, 8, 11, 21–25]. In this study, we examined factors associated with BSIs due to all species of non–C. albicans Candida, compared with BSIs due to C. albicans, among patients hospitalized in medical and surgical ICUs. In our study, an increased risk of BSI due to non-albicans Candida species was independently associated with duration of CVC use (P=.02) and proportion of days of fluconazole exposure (P=.003). Mean number of antibiotics received per day trended toward an increased risk of candidemia due to non-albicans Candida species, compared with C. albicans candidemia, but the difference did not reach statistical significance (P=.17). Duration of TPN was associated with a decreased risk of developing BSI due to non-albicans Candida species, compared with a C. albicans BSI (P=.009).
Potential risk factors for developing candidemia in surgical ICU patients identified in previous studies have included diabetes mellitus, prior surgical procedure, prior ICU admission, prolonged cardiopulmonary bypass time, mechanical ventilation, receipt of TPN, presence of CVCs, diarrhea, and bacteremia. However, these are risk factors for fungemia due to all species of Candida and are limited to patients hospitalized in a single type of ICU [26–28]. Because our study includes both medical and surgical ICU patients, our results apply to a more general critically ill adult population.
Our study findings contrast with those of Shorr et al. [29], who did not identify any differences in clinical factors between ICU patients infected with non-albicans Candida species and ICU patients infected with C. albicans. The study by Shorr et al. [29] was similar to ours in design and had comparable numbers of patients. However, the metrics used to measure risk factor exposure differed. Shorr et al. [29] only defined presence or absence of risk factor exposure (i.e., had the patient been previously given ⩾48 consecutive h of fluconazole in the past 90 days? [yes or no]) and did not account for exposure duration. It is biologically plausible to assume that the risk of candidemia accumulates with increased duration of exposure to a risk factor. In addition, Shorr et al. [29] did not adjust for patients' time at risk for exposure. In other words, was a patient exposed to more risk factors simply because he or she was present in the ICU for a longer period? Our study adjusted for both exposure duration and patients' time at risk for exposure, which allowed for a more sensitive measure of risk and the detection of several differences between patients infected with non-albicans Candida species and those infected with C. albicans.
We found fluconazole exposure to be a risk factor for the development of candidemia due to non-albicans Candida species (OR, 11.6; P=.009). In further analyses, when cases of candidemia due to non-albicans Candida species were divided into 2 groups according to expected patterns of fluconazole susceptibility, fluconazole exposure remained highly associated with each of the groups. The explanation for this association may be multifactorial, including any factor that would have resulted in a differential effect of fluconazole on different Candida species. In addition to fluconazole susceptibility, variability among Candida species includes site of colonization [30], age and type of host affected [31], and biofilm formation [32]. Production of biofilms is observed most frequently for C. tropicalis and C. parapsilosis (2 species that are generally thought to be azole susceptible) and less frequently for C. albicans [32]. Furthermore, fluconazole does not work well in biofilms [33]. This may lead to a discrepancy between in vitro susceptibility testing and clinical in vivo response to fluconazole. Additional evidence of a differential effect of fluconazole that is not likely to be solely explained by fluconazole susceptibility patterns is the association of prior fluconazole exposure with breakthrough candidemia due to non-albicans Candida species that are considered to be susceptible to fluconazole but not with candidemia due to C. albicans [8, 34–37].
We also found that the number of different antibiotics received per day, but not antibiotic type or duration of antibiotic therapy, was a risk factor for candidemia due to non-albicans Candida species. Nearly all patients with BSIs due to non-albicans Candida species (96%) and C. albicans BSIs (99%) received antibiotic therapy during their ICU stay. Because of these high rates of antibiotic use, it is not surprising that a difference in duration of antibiotic therapy was not detected between the 2 groups. These results contrast with those of Lin et al. [23], who found an association between exposure to vancomycin and piperacillin-tazobactam (but not fluconazole) and the development of C. glabrata or C. krusei candidemia.
In our study, duration of TPN exposure was associated with a decreased risk of developing a BSI due to non-albicans Candida species, compared with BSI due to C. albicans. Another way to interpret this finding is that TPN is a greater risk factor for C. albicans BSI than for BSI due to non-albicans Candida species. At initial examination, the association between TPN duration and a decreased risk of BSI due to non-albicans Candida species may seem unexpected, because TPN has been found to be a risk factor for candidemia overall [4, 5, 22, 26, 27]. TPN use may be a surrogate marker for increased severity of illness, because TPN is often associated with sicker, more debilitated patients. Not only is C. albicans more common than non-albicans Candida species, but C. albicans is also known to be more virulent than non-albicans Candida species [38–40]. Therefore, for critically ill patients in an ICU who are exposed to TPN, it is reasonable to note that the use of TPN is associated with an increased risk of C. albicans BSI, compared with candidemia due to non-albicans Candida species. Duration of TPN, however, cannot be used to distinguish between patients at risk of C. albicans BSI and patients at risk of BSI due to non-albicans Candida species.
In terms of our study's limitations, data were collected from urban, tertiary care hospitals, resulting in a sample that was presumably influenced by sicker patients with multiple comorbidities who may not be seen in a community setting. In addition, our study did not record severity of illness scores, which further limits comparison of our findings to other critically ill patient populations. Our observational study is also subject to bias of retrospective data collection, which prevented us from distinguishing between prophylactic and empirical antifungal therapy strategies. We also did not collect data on timing or appropriateness of treatment after the day when the positive blood sample was obtained. Although fluconazole exposure and other risk factors may be surrogate markers for causal factors, the aim of this study was to identify risk factors or associations with C. albicans BSI, compared with BSI due to non-albicans Candida species, and our goal was not to determine causal relationships. Although adjusted duration of corticosteroid therapy was included as a potential risk factor, it did not remain statistically significant in multivariate analysis, which contrasts with other studies [41–43]. We did not record steroid dose, which could have been more informative and may have lead to significance in multivariate analysis. We also did not distinguish between low-dose corticosteroids administered for replacement therapy for septic shock and higher doses of corticosteroids administered for other reasons. A strength of our study is that our data represented 2 centers and included patients from different ICU settings, including medical, surgical, cardiac, cardiothoracic, and neurosurgical ICUs and the trauma unit, which helps make our results more generalizable to critically ill patients. Furthermore, relative to other studies of candidemia, the sample size in our study was fairly large.
In summary, this report defines a number of factors that are independently associated with an increased risk of candidemia due to non-albicans Candida species, compared with C. albicans candidemia. Our findings suggest that receipt of fluconazole, presence of a CVC, and number of antibiotics are associated with an increased risk and that TPN exposure is associated with a decreased risk of BSI due to non-albicans Candida species, compared with BSI due to C. albicans. Until data from clinical trials are available, for ICU patients with these specific characteristics, clinicians might consider an empirical treatment strategy that initially uses an antifungal agent other than an azole that covers potentially resistant pathogens until the species can be identified. Critically ill patients without these specific characteristics, on the other hand, may safely initiate empirical fluconazole therapy. Additional studies to further define which antifungal agents are most effective for empirical therapy for different populations of patients hospitalized in ICUs, who are at high risk for such infection, are needed.
Acknowledgments
Financial support. National Institutes of Health (5 T32 AI055412-02 to J.C.) and Tufts–New England Medical Center.
Potential conflicts of interest. Y.G. has received research funding from Pfizer and Merck and has been a member of the speaker's bureau for Schering-Plough. A.W.K. has received research funding from and has served as a consultant and advisory board member for Pfizer. D.L. is currently employed by Bard Medical Division. S.H. has received research funding from Pfizer; has served as a consultant to Astellas, Domantis, Pfizer, and Schering-Plough; and has been on the speaker's bureau for Astellas, Enzon, Merck, Pfizer, and Schering-Plough. Y.C. has received grants, honoraria, travel support, and consulting fees from Basilea Pharmaceutica, Bioline Therapeutics, Cempra Pharmaceuticals, Johnson and Johnson, Merck, Neopharm, Pfizer, Wyeth, and XTL. All other authors: no conflicts.
- Received July 6, 2007.
- Accepted November 30, 2007.
- © 2008 by the Infectious Diseases Society of America
No comments:
Post a Comment