Evaluation of the predictive indices for candidemia in an adult intensive care unit

Introduction: To evaluate predictive indices for candidemia in an adult intensive care unit (ICU) and to propose a new index. Methods: A prospective cohort study was conducted between January 2011 and December 2012. This study was performed in an ICU in a tertiary care hospital at a public university and included 114 patients staying in the adult ICU for at least 48 hours. The association of patient variables with candidemia was analyzed. Results: There were 18 (15.8%) proven cases of candidemia and 96 (84.2%) cases without candidemia. Univariate analysis revealed the following risk factors: parenteral nutrition, severe sepsis, surgical procedure, dialysis, pancreatitis, acute renal failure, and an APACHE II score higher than 20. For the Candida score index, the odds ratio was 8.50 (95% CI, 2.57 to 28.09); the sensitivity, specifi city, positive predictive value, and negative predictive value were 0.78, 0.71, 0.33, and 0.94, respectively. With respect to the clinical predictor index, the odds ratio was 9.45 (95%CI, 2.06 to 43.39); the sensitivity, specifi city, positive predictive value, and negative predictive value were 0.89, 0.54, 0.27, and 0.96, respectively. The proposed candidemia index cutoff was 8.5; the sensitivity, specifi city, positive predictive value, and negative predictive value were 0.77, 0.70, 0.33, and 0.94, respectively. Conclusions: The Candida score and clinical predictor index excluded candidemia satisfactorily. The effectiveness of the candidemia index was comparable to that of the Candida score.

In the United States, the incidence of fungal infections increased 207% between 1979 and 2000.Candida spp. is the most common causative agent of such infections, accounting for 80% of cases (1) .Candidemia is considered to be the fi fth to tenth most common cause of bloodstream infection in patients treated in intensive care units (ICUs) (2) .
Colombo et al. (3) conducted a multicenter study in which the authors detected 712 cases of candidemia with an incidence density of 2.49 cases per 1,000 patients per day.Forty-six percent of the patients were staying in the ICU, whereas 39% were surgical patients.Compared with the United States, the incidence of candidemia in adult patients is higher in Brazil (0.46/1,000 versus 2.49/1,000, respectively).In both countries, candidemia represents the fourth most frequent cause of bloodstream infection (3) .Candidemia accounts for high mortality rates, typically when the onset of therapy is delayed for more than 48 hours (4) (5) .A positive blood culture for Candida spp.remains the gold standard for the diagnosis of candidemia (6) .However, this test cannot detect all cases of candidemia.Additionally, Candida spp.isolation may take too much time, thereby delaying effective antifungal therapy (7) .These diffi culties have motivated the search for risk factors and candidemia-associated conditions that could help predict this opportunistic infection and design preemptive antifungal therapy.
One study analyzed 2,890 patients who remained in the ICU for at least four days in nine hospitals in the United States and Brazil.The authors proposed a clinical prediction rule for invasive candidiasis in the intensive care setting.The main risk factors were as follows: use of antibiotics, a central catheter, immunosuppressants or corticosteroids; previous surgery; pancreatitis; parenteral nutrition; and dialysis (8) .
In 2009, León et al. (9) evaluated the use of the Candida score, an index that includes clinical data and multifocal colonization by Candida spp., to discriminate between Candida colonization and candidemia in non-neutropenic critically ill patients.The authors concluded that patients staying at the ICU for more than seven days but with a Candida score lower than 3 were at a 5% lower risk of candidemia (9) .
Posteraro et al. (10) employed the beta-D glucan assay, colonization index, and Candida score to assess 377 patients

METHODS
admitted to an ICU.The authors demonstrated that the Candida score provided the second best diagnostic yield: the negative predictive value was high (0.97), but the positive predictive value was low (0.57) (10) .
The need for early therapy to treat invasive infection by Candida spp.motivated the present study.We investigated the risk factors for candidemia and evaluated the use of the two mentioned predictors of candidemia in an adult ICU setting in Brazil.In addition, we analyzed a new predictor, the candidemia index.

Clinical setting, design, and study population
This prospective study aimed to compare a cohort of patients staying at the (ICU) of the University Hospital of the Ribeirão Preto Medical School, University of São Paulo, Ribeirão Preto, SP, Brazil (HC-FMRP-USP) for at least 48 hours.The study was conducted between January 2011 and December 2012.
The following data were collected for the participants: age, origin (clinical or surgical ward), underlying disease, central venous catheter, total parenteral nutrition, use of any type of dialysis, use of immunosuppressants (except corticosteroids), presence of neutropenia (<500 neutrophils/µL), previous surgery (any surgical procedure within the previous 30 days), presence of a site colonized by Candida (during ICU stay), intestinal perforation (up to seven days previously), APACHE II (Acute Physiology and Chronic Health Disease Classifi cation System II) score higher than 20, and Enterobacter bacteremia (up to seven days previously).
The blood culture samples were processed on an automated BD ® (Becton Dickinson, USA) system.The yeast isolated from the different samples was identifi ed as Candida spp.by means of the standard procedures of a mycological laboratory and through the use of the VITEK 2 ® system (bioMérieux, France).The latter instrument was also employed to evaluate the Candida susceptibility to fl uconazole.
The Ethics Committee of HC-FMRP-USP approved this study under Process number HCRP 6915/2010.

Defi nition of fungal colonization
Colonization was defi ned as Candida spp.isolation in some topography without causing disease in the host.We did not use active surveillance, but we evaluated the results obtained from cultures (urine, stool, tracheal secretion, and others) collected by an intensivist.

Calculation of the Candida score
The Candida score was calculated on the basis of the following score criteria: multifocal colonization (one score), surgery (one score), total parenteral nutrition (one score), and signs of severe sepsis (two scores), which amounted to fi ve (5) scores (11) .

Calculation of the clinical predictor index
The participants were classifi ed as being at positive risk for candidemia when they were receiving systemic antibiotics or had a central venous catheter as well as at least two of the following risk factors: parenteral nutrition (minimum of three days), dialysis (minimum of three days), surgery (within seven days previously), pancreatitis (within seven days previously), corticosteroids (minimum of three days), and any other immunosuppressants (minimum of seven days).All of the patients were classifi ed (positive or negative risk) according to these criteria (8) .

Calculation of the candidemia index
This study evaluated an alternative criterion to predict candidemia using 16 factors assessed: each factor or condition that was present in the patient was awarded a score of one (total = 16).Patients were classifi ed after 48 hours in the ICU.Only one classifi cation was performed per patient according to the inclusion criteria.Patients were excluded if they had stayed at the ICU less than 48 hours or were hematological patients who had not developed neutropenia (<500 neutrophils/uL) during hospitalization.The risk factors analyzed may have fi rst presented in the ICU or in the patient's original ward.More specifi cally, the following risk factors were analyzed: parenteral nutrition (minimum of three days), surgery (within the previous seven days), severe sepsis (with organ dysfunction), broad-spectrum antibiotics (previous use, up to 30 days, of fourth-generation cephalosporins, carbapenems or glycopeptides), central venous catheter (fi tted during hospital stay), immunosuppressants (minimum of seven days), dialysis (minimum of three days), pancreatitis (within the previous seven days), corticosteroids (minimum of three days), acute renal failure (during hospital stay), intestinal perforation (during hospital stay), nosocomial infection (during hospital stay), APACHE score >20, Enterobacter bacteremia (during hospital stay), fungal colonization (evaluated during the ICU stay), and neutropenia (during the ICU stay).The receiver operating curves (ROC) was constructed to establish the best cutoff point for this index to predict candidemia.

Statistical analysis
To verify the association between the variables, the Fisher's exact test was employed.This procedure was conducted using the software SAS® 9.2 (Statistical Analysis System, USA) and PROC FREQ.Comparison between the Candida score groups and the candidemia index relied on the Mann-Whitney test for independent samples.

Participant characteristics
This study included 114 adult patients with a mean age of 57.8 years.Clinical patients represented 57.9% of the study population; the mean APACHE II score was 24.3, and the mean death risk was 49%.Blood culture revealed 18 cases of candidemia, whereas 96 patients did not present with this condition (Table 1).The length of stay at the adult ICU was approximately the same for all the participants regardless of invasive infection by Candida, but the mortality was higher (61.1%) in the group with candidemia.With respect to the underlying disease, 11 (61.1%)patients with candidemia had neoplasia.Six (54.5%) and fi ve (45.5%) of the patients with candidemia had hematological and solid organ malignancies, respectively.
With respect to sensitivity, all samples of Candida albicans, Candida tropicalis, and Candida parapsilosis were susceptible to fluconazole.However, 7 Candida krusei samples were resistant to fluconazole (63.6%): 3 urine culture samples, 2 tracheal secretion samples, 1 blood culture sample and 1 catheter tip sample.All of the samples of Candida glabrata were sensitive to fl uconazole.The sensitivity and resistence to fl uconazole were 94.34% and 5.66%, respectively.
Table 3 lists the application of the Candida score and the clinical predictor index to all of the patients included in the study.The Candida score was statistically signifi cant for the presumptive diagnosis of candidemia (p < 0.01), and the odds ratio was 8.50 [95% confi dence interval (CI)], from 2.57 to 28.09); the sensitivity, specifi city, positive predictive value, and negative predictive value were 0.78, 0.71, 0.33, and 0.94, respectively.
The clinical predictor index was also statistically signifi cant for the presumptive diagnosis of candidemia (p < 0.01), and the odds ratio was 9.45 (95% CI, from 2.06 to 43.39); the sensitivity, specifi city, positive predictive value, and negative predictive value were 0.89, 0.54, 0.27, and 0.96, respectively (Table 3).

Performance of candidemia index compared
with the Candida score The ROC curve (Figure 1) aided the estimation of the best cutoff point, 8.5, for the candidemia index.The sensitivity,   specifi city, positive predictive value, and negative predictive value were 0.77, 0.70, 0.33, and 0.94, respectively.The area under the curve (AUC) was 80.8, with a 95% confi dence interval (72.6; 88.9).The total score statistically correlated (p < 0.01) with cutoff point 3 of the Candida score (Figure 2).

DISCUSSION
The main achievement of this study was to validate predictive indices that are particularly useful to exclude candidemia in patients admitted to the ICU of a tertiary hospital located in the interior of Brazil.
With respect to severity, the group with candidemia exhibited a higher mean APACHE II score than the group without candidemia.Therefore, patients who acquired candidemia exhibited more critical conditions, which could imply more interventions, such as the frequent use of invasive procedures.Fraser et al. (12) demonstrated that mortality in candidemia patients is directly related to the APACHE II score (12) .
In the present study, neoplasia was the most prevalent underlying disease among patients with candidemia.Han et al. (13) reported similar fi ndings, in which candidemia patients had more malignant neoplasms compared to non-candidemia patients (13) .When Ostrosky-Zeichner et al. (8) validated the clinical predictor index for candidemia, they also identifi ed neoplasm as a risk factor (8) .
Analysis of the patients' progression revealed a mortality of 61.1% among subjects with candidemia.Various studies have

Candidemia index
also verifi ed a general mean mortality rate and an attributed mortality rate of approximately 60% and 40%, respectively, among patients with candidemia (9) (14) .
Among the risk factors investigated in the present study, we found that parenteral nutrition, severe sepsis, surgical procedures, dialysis, pancreatitis, acute renal failure, and an APACHE II score higher than 20 were associated with candidemia (Table 3).Blumberg et al. (15) identifi ed parenteral nutrition [relative risk (RR) = 3.6], acute renal failure (RR = 4.2), and previous surgical procedures (RR = 7.3) as risk factors for candidemia (15) .Cheng et al. (16) and Mokadas et al. (17) also demonstrated that renal failure was an independent risk factor for the onset of candidemia (16) (17) .
In a prospective cohor1t study, León et al. (11) also identifi parenteral nutrition, severe sepsis, and multifocal Candida colonization as risk factors (9) (11) .In 2009, this same team validated the Candida score and reported a relative risk of 5.9 for the occurrence of candidemia when the total score was equal to or higher than 3. Here, the Candida score provided similar results to those obtained in the validation conducted by those authors (9) (11) : sensitivity 0.78 vs 0.77, specifi city 0.71 vs 0.77, positive predictive value 0.33 vs 0.13, and negative predictive value 0.94 vs 0.97.
In the present study, the incidence of candidemia was low (5%) when the Candida score was less than 3 and 33% when the Candida score was equal to or higher than 3. Leroy et al. (18) reported incidences of candidemia of 0%, 17.6%, and 50% in patients with scores of 2 or 3, 4, and 5, respectively (18) .These data confi rmed that the Candida score successfully excluded candidemia at the bedside.
Ostrosky-Zeichner et al. (8) described parenteral nutrition, pancreatitis, dialysis, previous surgery, and the use of antimicrobials, central venous catheters, immunosuppressants, or corticosteroids as risk factors for candidemia (8) .The clinical predictor index proposed by those authors was evaluated for the patients in the present study and provided increased sensitivity and reduced specifi city.The parameters obtained by Ostrosky-Zeichner et al. (8) and this study for the clinical predictor index were sensitivity 0.43 vs 0.89, specifi city 0.90 vs 0.54, positive predictive value 0.01 vs 0.27, and negative predictive value 0.97 vs 0.96, respectively.
We assessed a newly proposed index that considered the total score (candidemia index) of each patient included in the study; we adopted a cutoff point of 8.5.In these conditions, the candidemia prediction was similar to that achieved with the Candida score (9) .We verifi ed a high correlation between the candidemia index proposed here and the Candida score.The candidemia index must be validated through studies involving a larger number of participants and evaluating different weights for each risk factor.
In conclusions, our data determined that the risk factors associated with candidemia were similar to those observed in hospitals of other geographic regions.The Candida score and the clinical predictor index performed well, which suggested that they might be useful in the clinical setting, especially to exclude candidemia.The candidemia index, a new predictive index for candidemia based on risk factors, was comparable to the Candida score.

FIGURE 1 -
FIGURE 1 -ROC curve for the total score (candidemia index) obtained from the patients according to the number of risk factors for candidemia.AUC: area under the curve; CI: confi dence interval; ROC: receiver operating curves.*Value with best sensitivity and specifi city.

FIGURE 2 -
FIGURE 2 -Box plot of the relationship between the total score (candidemia index) and the candida score.+ Statistical signifi cance at p < 0.05.

Gaspar
GG et al. -Evaluation of the predictive indices for candidemia

TABLE 1 -Clinical and demographic characteristics of the study population according to the presence or absence of candidemia among patients admitted to the adult intensive care unit.
APACHE II: acute physiology and chronic health evaluation II.

TABLE 2 -Risk factors for candidemia in patients admitted to the adult intensive care unit.
APACHE II: acute physiology and chronic health evaluation II.*Defi ned as sepsis with organ dysfunction.**Excluding the use of corticosteroids.+ Statistical signifi cance at p < 0.05.

TABLE 3 -Prediction of candidemia by the Candida score and clinical predictor index in patients admitted to the adult intensive care unit.
CI: confi dence interval .+ Statistical signifi cance at p < 0.05.