Take home message

The epidemiology of candidemia is changing with the constant evolution of medical and surgical care. In this study, we show that the risk of candidemia depends on exposure to different antibiotics and/or medical procedures in ICU and non-ICU patients, highlighting the need for setting-specific risk assessment scores.

Introduction

Candida spp. are the third most common microorganisms responsible for health-care-related bloodstream infections [1]. The incidence of candidemia has increased by 50% over the last decade worldwide and ranges between 2.4/100000 and ~ 15/100000 individuals, depending on the country and clinical setting [2,3,4,5,6]. Despite significant progress in antifungal treatment options, candidemia is still associated with an overall crude mortality rate ranging between 40 and 60% [4, 7,8,9,10,11]. Attributable mortality ranges from 5% to 49% [12,13,14], depending on the control group considered and the underlying comorbidities, the impact of nosocomial infections being known to be greater in less sick population, and so probably less important in ICU patients [15]. Prompt initiation of appropriate antifungal therapy is crucial to improve the chances of survival [16]. However, blood cultures for yeasts lack sensitivity and need prolonged incubation (> 24 h). As a consequence, antifungal drugs are often prescribed either prophylactically, pre-emptively, or empirically in high-risk patients [17]. The resulting overuse of antifungal drugs may lead to the emergence of Candida species that are resistant to azoles and/or echinocandins [5, 18,19,20].

Few studies used a matched case-control design to assess risk factors for candidemia [21,22,23,24,25]. Unmatched studies identified factors such as a central venous catheter (CVC), prior surgery, broad-spectrum antibiotic therapy, or total parenteral nutrition (TPN) which are present in a large number of hospitalized patients [22, 23, 26,27,28]. Furthermore, most studies were performed either inside or outside intensive care units (ICUs) and a few of them allowed for differential analyses according to both settings [24]. This prospective, multicenter, matched case-control study aims to assess the risk factors associated with candidemia in high-risk groups of patients in both the ICU and non-ICU settings.

Materials and methods

Study design and patients

This multicenter, international, prospective, matched case-control study was carried out in five university hospitals (Lille, France; Lausanne, Geneva, Bern, and Basel, Switzerland) and a large teaching hospital (St. Gallen, Switzerland) contributing to the Fungal Infection Network of Switzerland (FUNGINOS)—and ALLFUN networks between July 2013 and March 2017. Patients were included if they were > 18 years old with at least one blood culture positive for Candida spp. Matched controls (up to three per case) were selected by local investigators for each case. Matching criteria included age (+/− 5 years), hospital ward, duration of hospital stay (time from hospital admission to candidemia in each case was matched to a length of hospitalization at least equal for the corresponding control; most controls remained hospitalized after their inclusion, they were followed-up to ensure that they did not develop candidemia), and the type of surgery in case of surgical procedure. Patients with a history of intravenous drug abuse were excluded from the study as they usually have a clinical risk profile that is different from other candidemic patients.

Laboratory tests

Two automated blood culture systems were used during the study period: Bactec™ (Becton Dickinson, Sparks, Maryland, USA) and Bact/Alert®3D (bioMérieux, Marcy l’Etoile, France). Yeasts isolated from blood cultures were identified by MALDI-TOF mass spectrometry (Microflex Mass Spectrometer, Bruker Daltonics GmbH, Bremen, Germany) as described previously [29]. Isolates with MALDI-TOF score less than 1.7 were subsequently identified by molecular methods, as reported previously [30].

Data collection and definitions

Demographic characteristics and underlying medical conditions were recorded systematically for each case and matched controls in a secured electronic case report form (eCRF). Corticosteroid use was defined by the use of > 20 mg prednisone-equivalent daily for > 10 days before positive blood cultures. Clinical conditions and risk factors within 2 weeks prior to candidemia (or a matched time in controls) were also recorded, including the presence of intravenous and urinary devices, TPN, mechanical ventilation (for > 24 h), renal replacement therapy, and use of gastric acid secretion inhibitors. The use of antibacterial and antifungal drugs within 4 weeks before candidemia (or equivalent time in controls) was also recorded. Whenever available, Candida colonization index and Candida score were recorded by using the method described by Pittet et al. [31] and Léon et al. [22, 27]. We defined ICU population as patients hospitalized in ICU at the time of candidemia and conversely for non-ICU population.

Statistical analysis

Statistical analyses were performed using the Stata software (v15.1; College Station, TX, USA). Factors associated with candidemia and mortality were analyzed by using univariate and multivariate conditional logistic regression models. A backward stepwise logistic regression was used to select variables entered in the multivariate models, using a cutoff p value of 0.10. New scores to predict the risk of candidemia were developed for patients in and outside the ICU. Scores were obtained by rounding the β-coefficients. Receiving operating characteristic (ROC) curves were drawn using rocreg implemented in Stata®, after adjustment for matching covariates [32]. Test efficiencies were calculated using the dtroc softwares (Stata®). The best cutoff point was established according to standard methods (Youdden’s approach to determine the cutoff with the best compromise between sensitivity and specificity; the method of Zweig and Campbell, maximizing efficiency) [33, 34] by using cutpt (Stata®).

Results

Study population

The study included 192 patients with candidemia and 411 controls matched for age, hospital duration stay, ward, and type of surgery in case of surgery. Patients were almost equally distributed between surgical (56%) and medical wards (44%) and between non-ICU (53%) and ICU (47%). Median age was 63 years [52–74] and approximately two-thirds of patients were male. Candidemia occurred within a median duration of 16 days (interquartile range 5–27) after hospital admission. Candida albicans was the most commonly reported species (61%), followed by Candida glabrata (16%), Candida parapsilosis (9%), Candida tropicalis (3%), Candida krusei (3%), and other/mixed species (8%).

Risk factors for candidemia

Univariate and multivariate analyses of risk factors for candidemia according to hospital setting are shown in Table 1 and in Table 2, respectively. Independent risk factors for candidemia in the whole population included central venous catheter (OR = 6.74, 95% confidence interval (CI) 2.96–15.4, p < 0.001), TPN (OR = 3.92, 95%CI 2.28–6.73, p < 0.001), previous septic shock (OR = 2.29, 95%CI 1.33–3.96, p = 0.003), exposure to nitroimidazoles (OR = 2.16, 95%CI 1.11–4.21), and renal replacement therapy (OR = 2.16, 95%CI 1.11–4.21, p = 0.02).

Table 1 Demographic and clinical characteristics of patients with candidemia and matched controls inside and outside intensive care units
Table 2 Independent risk factors associated with candidemia according to hospitalization inside and outside intensive care units

Independent risk factors for candidemia within the ICU population included TPN (OR = 6.75, 95%CI 2.89–15.7, p < 0.001), acute kidney injury (OR = 4.77, 95%CI 1.94–11.8, p < 0.001), heart disease (OR = 3.78, 95%CI 1.09–13.1, p = 0.006), previous septic shock (OR = 2.39, 95%CI 1.14–5.01, p = 0.02), and exposure to aminoglycosides (OR = 2.28, 95%CI 1.01–5.13, p = 0.05).

Independent risk factors for candidemia within the non-ICU population included CVC (OR = 9.77, 95%CI 3.72–25.7, p < 0.001), TPN (OR = 3.29, 95%CI 1.52–7.13, p = 0.003), exposure to glycopeptides (OR = 3.31, 95%CI 1.33–8.23, p = 0.04), and to nitroimidazoles (OR = 3.12, 95%CI 1.07–9.11, p = 0.04).

Predictive scores for candidemia based on the aforementioned risk factors were developed for both ICU and non-ICU patients (Fig. 1, panel A1 and A2, respectively). The area under the curve (AUC) was 0.768 for ICU patients and 0.717 for non-ICU patients. The optimal cutoff value for the best compromise between sensitivity and specificity was ≥ 4 for ICU patients (sensitivity = 69%, and specificity = 70%) and ≥ 2 for non-ICU patients (sensitivity = 83% and specificity = 49%). Considering a method maximizing efficiency, the optimal cutoff for a better specificity was ≥ 5 for ICU patients (sensitivity = 43%, specificity = 88%) and ≥ 4 for non-ICU patients (sensitivity = 51% and specificity = 81%).

Fig. 1
figure 1

Risk scores for candidemia. Scoring values assigned to each variable (A1 and A2), resulting ROC curves with adjusted areas under the curve (aAUCs, B1 and B2) and single risk performance values (C1 and C2) are shown for patients inside and outside ICU, respectively. Se, Sp, LR+, and LR− stand for sensitivity, specificity, positive and negative likelihood ratios, respectively. The number of patients included in the calculation of score may be lower than the total number of patients due to missing co-variables in some individual patients

Risk factors of mortality

Univariate and multivariate analysis of risk factors for death in candidemic patients according to hospital setting are shown in Table 3 and in Table 4, respectively. Independent risk factors for death in the whole population included septic shock (OR = 6.80, 95%CI 2.93–15.8, p < 0.001), acute kidney injury (OR = 5.62, 95%CI 2.44–12.9, p < 0.001), and the number of antibiotics (OR = 1.43, 95%CI 1.16–1.77 per unit, p < 0.001). Age tended to be associated with death (p = 0.06). Independent risk factors for death in ICU patients included septic shock (OR = 4.09, 95%CI 1.72–14.0, p = 0.003), acute kidney injury (OR = 3.45, 95%CI 1.21–9.90, p = 0.02), and the number of antibiotics to which patients were exposed before candidemia (OR = 1.37, 95%CI 1.06–1.75 per unit, p = 0.02). Independent risk factors for death in non-ICU patients included acute kidney injury (OR = 11.9, 95%CI 2.47–57.7, p = 0.002) and septic shock (OR = 8.70, 95%CI 2.26–33.5, p = 0.002).

Table 3 Risk factors for death in candidemic patients, according ICU vs non-ICU setting
Table 4 Independent risk factors of death associated with all-cause death in candidemic patients according to the ICU vs non-ICU hospital setting

Discussion

This prospective, multicenter, matched case-control study was designed to analyze risk factors for candidemia in both ICU and non-ICU patients. The study included the largest number of candidemic patients reported from a case-control study in the ICU [25] and the second largest sample size for a case-control study outside the ICU [21]. Different risk factors for candidemia were identified in both settings, allowing for targeted risk factor selection.

Because invasive candidiasis is a rare clinical event, previous studies have included cases irrespective of the presence or absence of candidemia [22,23,24, 27]. Non-candidemic patients can represent up to 30% of cases in some studies [22, 24]. The term “invasive candidiasis” is applied to very differently defined clinical conditions. Some of these, such as post-surgical intra-abdominal candidiasis, require a complex diagnostic approach with clinical and microbiological expertise [35], while others, such as candidemia, represent a clear-cut phenotype. In order to maximize case homogeneity and minimize the risk for misclassification, we considered only patients with candidemia in the present study. Furthermore, we used a matched case-control design, with matching criteria similar to those used in the seminal paper by Wey et al. [25], adding a more stringent matching for the type of surgery. A novelty of the present study is the application of a matched case-control design in ICU patients. The matching criteria aimed at separating risk factors that are specific for candidemia from those that result from prolonged hospitalization [25].

Overall, the study confirms the well-established risk factors for candidemia, such as total parenteral nutrition (the most robust one, which was identified in all studies [10, 21,22,23,24,25, 27]), central venous catheter [10, 23,24,25, 28], septic shock [21, 22], kidney failure, or renal replacement [10, 23, 25], as well previous exposure to antibiotics (without class specification) [21, 23,24,25]. The study also highlights the specific risk factors for candidemia that emerge for the ICU and the non-ICU settings, as illustrated by specific patterns of antibiotic exposure, as well as clinical features or medical equipment. For instance, CVC was an independent risk factor for candidemia outside the ICU, probably reflecting its very frequent use (> 90% of patients) in ICU, making it non-discriminant for the determination of the risk of candidemia this setting [22]. In contrast, septic shock was associated with candidemia solely inside the ICU, in accordance with the two studies by Léon et al. [22, 27], reflecting the fact that most patients with such complication are managed in this setting. The other clinical features associated with candidemia solely among ICU patients included heart failure and kidney injury which not previously reported in this setting.

One of the most striking findings of this study was the different patterns of antibiotic exposure associated with candidemia in ICU and non-ICU patients. Glycopeptides and nitroimidazoles were associated with candidemia only outside the ICU. The frequent use of these drugs in the ICU may explain the lack of association in this specific setting. This finding is consistent with a recent study in patients on internal medicine wards, in which glycopeptides were found to be an independent risk factor for candidemia [21]. As intraabdominal bacterial infections were associated with candidemia in the ICU population in univariate analysis (but not in multivariate one), we have forced this variable in the multivariate models for the whole population and the non-ICU one to check for bias. The association between candidemia and glycopeptides/nitroimidazoles remains significant so that these classes of antibiotics can be considered as independent from intraabdominal bacterial infections. In contrast, aminoglycosides were an independent risk factor for candidemia solely in the ICU. These drugs may represent a supplementary risk factor for developing candidemia among ICU patients, who are exposed to multiple other classes of antibiotics (including drugs active against Gram-negative anaerobic bacteria) and/or to glycopeptide antibiotics. Because control matching was performed on a center basis, the associations with antibiotics are not likely to reflect any differences in center’s antibiotic stewardship or empirical treatment strategies.

Candida colonization was previously reported as a risk factor for candidemia in some studies [22, 25, 27], but not in others [21, 23, 24]. Colonization was not systematically tested in all patients, thereby limiting the statistical power to detect an association with candidemia. The different practices to monitor Candida colonization among centers due to logistic and financial issues may limit its universal use to assess the risk of candidemia. On the other hand, Candida colonization, if systematically monitored over time during prolonged hospitalization, may become too frequent to be a discriminant predictor [36]. Corticosteroids and other immunosuppressive drugs were not associated with candidemia in the present study, neither in ICU nor in non-ICU patients. Corticosteroids were inconstantly associated with candidemia in previous studies, possibly due to the lack of standard definitions for high-risk corticosteroid dose and duration of exposure [21,22,23,24].

Both the non-ICU and ICU predictive scores for candidemia in this study can be used with relative low cutoff values. The high negative-predictive values associated with low cutoffs can be useful to identify patients in whom the occurrence of candidemia is unlikely, thereby avoiding the use of unnecessary antifungal prophylaxis or empirical/pre-emptive therapy [22, 27]. Alternatively, high positive-predictive values associated with higher cutoffs are applied in other studies for selecting patients who might benefit from empirical/pre-emptive antifungal therapy [21, 23]. In our study, the accuracy and the compromise between sensitivity and specificity is better for the ICU score than for the non-ICU score. The score in the ICU setting could be used both to exclude candidemia (low cutoff) or to detect candidemic patients (high cutoff). The scores should be validated and evaluated in a validation cohort.

This study extends the list of risk factors for candidemia that exert a strong influence on the intestinal microbiota. The gut is the most frequent portal of entry for invasive infection due to Candida spp. [37], as a key locus for host-pathogen interactions [38] and a major determinant for the transition from colonization to infection [39]. In mice models, TPN and subsequent enteral deprivation lead to important modifications in the gut microbiota (with a shift of the predominance of Gram-positive Firmicutes to Gram-negative Proteobacteria), alteration in the barrier function of epithelial cells [40], and intestinal inflammation [40, 41]. In mice, antibiotic administration is increasingly shown to exert important and long-lasting alterations on the gut microbiota, which can induce proliferation of pathogenic microorganisms [42]. Administration of drugs such as carbapenems [43], fluoroquinolones [44], and glycopeptides [45], this last one being recognized as independent risk factors for candidemia in the present study, has been associated with increased Candida gut colonization in mice, as a probable result of altered relative proportions of anaerobic and aerobic bacteria in the microbiome.

The results from this study are strengthened by a large sample size, with the largest collection of candidemia cases from ICU in a case-control study today and a prospective case-control study design. Yet, control matching implies the use of conditional regression models, which limits statistical power. Furthermore, the number of controls per case is smaller than in a cohort study, thereby limiting predictive score performance. The ICU setting and surgery were used as matching criteria and thus were not assessable as risk factor in this study. While our study suggests that risk assessment and scoring need to account for the hospital setting (ICU versus non-ICU), larger studies allowing for scores in even more specific groups of patients (such as medical, surgical, onco-hematological patients) would further improve risk prediction.

Conclusion

We show that risk factors for candidemia are different among patients hospitalized inside and outside ICUs. Specific patterns of antibiotic exposure are emerging as novel risk factors for candidemia. These include aminoglycosides for patients hospitalized within the ICU and glycopeptides and nitroimidazoles for patients hospitalized outside the ICU. Weighted scores predictive of candidemia can be built based on these risks. An improved prediction of the risk of candidemia may contribute to guide targeted preventive and therapeutic antifungal strategies.