Introduction

Invasive candidiasis (IC) is a spectrum of diseases including candidemia and deep-seated candidiasis [1, 2•]. Critically ill patients are among those at the highest risk of developing IC, and approximately one-third of all candidemia episodes are diagnosed in intensive care units (ICUs) [3, 4].

Candidemia is the most common form of IC, with a reported prevalence of 6.9 episodes per 1000 ICU patients [5]. With regard to deep-seated candidiasis, intra-abdominal candidiasis (IAC) accounts for most cases [6]. Other forms of deep-seated candidiasis observed more rarely in ICU patients include hematogenous disseminated disease such as hepatosplenic, ocular, cardiac, central nervous system, bone and renal candidiasis, which are seen more frequently in patients with prolonged candidemia, and/or in immunosuppressed patients [7].

Early antifungal treatment (before etiological diagnosis by classical culture-based methods) could be crucial in critically ill patients with suspected IC, since delays in treatment have been associated with increased mortality in those who truly have IC [8,9,10,11,12]. On the other hand, overtreatment may be harmful, exposing patients with a clinical presentation compatible with IC, but actually caused by other organisms (usually bacteria), to unnecessary drug toxicities and increased risk of emergence of antifungal resistance in commensal Candida isolates. In the attempt to find the right balance between overtreatment and undertreatment of critically ill patients with suspected IC, various early antifungal strategies have been proposed: (i) risk score-based empirical antifungal therapy; (ii) diagnostic-driven antifungal therapy based on the results of rapid non-culture-based tests; and (iii) pre-emptive therapy based on results of non-culture based tests performed as screening measure in patients at risk of IC but without manifest signs and symptoms of infection.

In this narrative review, we discuss recent literature regarding early antifungal therapy in critically ill patients, focusing in particular on the current role of empirical antifungal treatment.

Methods

A PubMed search was conducted for English language articles published before 1 December 2023, using various combinations of the key words “invasive candidiasis”, “critically ill” and “empirical treatment”, with particular focus on the last 5 years. Articles pertinent to the topic were selected for full text review based on authors’ judgment, and references of selected articles were also screened for retrieving further pertinent literature, focusing in particular on results from randomized controlled trials (RCTs). Eventually, the present narrative review was structured in the following sections: (i) a brief background regarding epidemiology, prognostic impact, and diagnosis of IC in the ICU; (ii) a discussion of current literature regarding early treatment of IC in the ICU; (iii) conclusions (including discussion of previous sections).

Background

Epidemiology and Prognostic Impact of Invasive Candidiasis in the ICU

IC remains frequent and associated with poor outcomes in the ICU, as reported in a large multinational, multicenter, retrospective study conducted in 23 ICUs across different European countries (EUCANDICU project), in which the cumulative incidence of IC was 7.07 episodes per 1000 ICU admissions with a crude 30-day mortality of 42% [13]. The non-mutually exclusive cumulative incidences of candidemia and IAC were 5.52 and 1.84 episodes per 1000 ICU admissions, respectively [13]. In the observational, prospective, multicenter EUROBACT study, conducted in 162 ICUs in 24 countries; 28-day mortality of candidemia was as high as 41% [14].

The epidemiology of IC has evolved in recent years, with a progressive shift from C. albicans to non-albicans Candida spp. observed globally, although it should be acknowledged that the shift was less certain or marked when the analysis was limited to ICU wards in a recent meta-analysis and meta-regression [15,16,17,18]. In the EUROBACT study, the relative frequency of C. albicans as causative agent of candidemia was 57%, followed by C. glabrata and C. parapsilosis, while in the more recent EUROBACT-2 study, C. albicans was responsible for 40% of fungemia episodes, and 58% were caused by non-albicans Candida species [14, 19•]. In the prospective, observational EPIC III study, involving 8135 ICU patients with suspected or proven infection from 1150 centers in 88 countries, C. albicans was isolated from 10% of patients with positive microbiological samples, while non-albicans species were 5% of the isolates [20].

Regarding antifungal resistance, in the SENTRY Antifungal Surveillance Program, including 20,788 Candida spp. isolates from different continents, fluconazole resistance in the most frequent Candida species ranged from 0.3% (C. albicans) to 8.1% (C. glabrata) [15]. Resistance to echinocandins was uncommon among isolates of C. albicans (0.0–0.1%), C. parapsilosis (0.0–0.1%), C. tropicalis (0.5–0.7%), and C. krusei (0.0–1.7%) [15]. Echinocandin-resistant C. glabrata has been more frequently reported from the USA (reported prevalence up to 12% in some centers) [21, 22], with cross-resistance to fluconazole in up to one-third of isolates [23], than from Europe [15]. In addition, the emergent pathogen C. auris has caused outbreaks in ICU wards worldwide, posing a complex challenge with regard to IC management [24•, 25]. Indeed, C. auris isolates are usually resistant to fluconazole and they show variable amphotericin and echinocandin susceptibility, with concomitant resistance to azoles, echinocandins, and polyenes having been reported [26,27,28].

Brief Diagnostic Considerations

Microscopic examination is rapid and can be helpful but a negative result does not exclude IC [29]. Cultures are the diagnostic gold standard but are positive in only 50–70% of cases of candidemia, and even less frequently in IAC (5–20%) [30, 31]. Furthermore, it can take several days before Candida is identified at species level and antifungal susceptibility test results are available. Recently, non-culture-based blood tests have been developed for detection of components of the fungal cell wall by immunoassays, DNA by PCR, and antibodies by serology. Their role in guiding diagnostic-driven early antifungal therapy is discussed below.

Early Antifungal Therapy of IC in the ICU

Risk Score-Based Empirical Antifungal Therapy

Numerous risk factors for IC have been identified, for example high Acute Physiology and Chronic Health Evaluation II (APACHE II) score, diabetes mellitus, renal insufficiency, surgery (especially abdominal surgery), pancreatitis, the use of broad-spectrum antibiotics, parenteral nutrition, hemodialysis, mechanical ventilation, presence of central vascular catheters, previous Candida colonization, and therapy with immunosuppressive agents [3, 32,33,34,35]. These risk factors have been condensed into various prediction scores (or risk scores) validated in clinical studies [36,37,38,39,40,41,42,43]. Risk scores can be considered as early diagnostic tools when evaluating empirical antifungal therapy in ICU at the bedside of patients with signs and symptoms consistent with IC. However, one challenge is that most risk factors for IC are also common among ICU patients without IC. Accordingly, risk scores usually demonstrate high negative predictive value (NPV), while their positive predictive value (PPV) is usually low [44].

Risk scores based on the burden of Candida colonization were initially proposed in the twentieth century to guide empirical antifungal therapy [38, 45, 46], while many other scores based only on clinical variables or on a combination of clinical variables and Candida colonization have been proposed subsequently. Some well-known risk scores are available in the literature. For example, Ostrosky‐Zeichner and colleagues developed a clinical prediction rule for IC associated with a 10% risk of IC, and 97% NPV [40]. Such a risk is defined in the presence of the following items: any systemic antibiotic OR presence of a central venous catheter AND at least two of the following: total parenteral nutrition, any dialysis, any major surgery, pancreatitis, any use of steroids or use of other immunosuppressive agents [40]. Paphitou and colleagues also developed a prediction rule conferring a 20% risk of IC to any ICU patient with at least one among diabetes mellitus, new onset hemodialysis, and use of total parenteral nutrition, plus ICU stay longer than 4 days, use of broad-spectrum antibiotics, and no receipt of antifungals [39]. According to the Candida score, developed by Leon and colleagues, a cumulative score of > 2.5 points was suggested as a possible cutoff for initiating empirical antifungals (based on a risk ratio of 7.35). Items defining the score included the following: multifocal Candida colonization (1 point); total parenteral nutrition (1 point); surgery as the reason of ICU admission (1 point); and clinical symptoms of severe sepsis (2 points) [41]. Guillamet developed a score based on clinical variables for predicting the risk of candidemia in a large retrospective cohort of patients with severe sepsis or septic shock [47]. The authors identified administration of total parenteral nutrition (+ 2 points), prior antibiotic exposure (+ 2 points), transfer from an outside hospital (+ 1 points) or admission from a nursing home (+ 2 points), mechanical ventilation (+ 1 points), and presence of a central venous catheter (+ 2 points) as independent predictors of candidemia, while the lung as the presumed source of sepsis was conferred -6 points [47]. Overall, the risk reached 43% for + 8 points [47]. In a prospective cohort study of nonneutropenic patients admitted to the ICU for ≥ 72 h, Playford and colleagues identified the following factors to be independently associated with an increased risk of IC: emergency gastrointestinal or hepatobiliary surgical procedure; ICU admission from operating theater, emergency department, or another hospital; presence of an uncoated central venous catheter; total parenteral nutrition; high-dose corticosteroids; prior antibiotic exposure; and prior Candida colonization of throat or urine [48]. The authors identified three patient cohorts based on two threshold scores: patients at high risk (PPV 11.7%); patients at intermediate risk (PPV 1.46%); and patients at low risk (PPV 0.24%) [48].

Regarding current evidence from RCTs on the use of an empirical treatment approach to IC in critically ill patients, EMPIRICUS was a multicenter double-blind randomized controlled trial (RCT) conducted in 260 nonneutropenic critically ill patients with ICU-acquired sepsis, multiple Candida colonization, multiple organ failure, and exposed to broad-spectrum antibacterial agents that compared empirical micafungin vs. placebo [49]. The primary time-to-event endpoint was invasive fungal infection-free survival at day 28. Of note, this endpoint highlights the frequent dichotomy observed in RCTs in this field regarding the aims of the administered antifungal therapy: (i) to improve survival of patients with signs and symptoms of infection (in this case, ICU-acquired sepsis) as empirical therapy; (ii) to prevent development of future IC events, as prophylactic therapy. Overall, despite a trend toward a favorable effect of empirical antifungal therapy, invasive fungal infection-free survival at day 28 was not significantly different in the micafungin arm vs. placebo arm, both in the entire study population (hazard ratio [HR] 1.35, 95% confidence interval [CI] 0.87–2.08]) and in the subgroup of patients with high (> 8 points) sequential organ failure assessment (SOFA) score (HR 1.69,95% CI 0.96–2.94) [49]. The INTENSE trial was another RCT evaluating early antifungal therapy with micafungin vs. placebo in critically ill patients with intra-abdominal infection requiring emergency surgery [50]. Of note, antifungal therapy was defined as “pre-emptive” in this study since the primary outcome was incidence of IC, although from this standpoint it could be considered as prophylaxis rather than preemptive therapy (i.e., antifungal agents to prevent future IC without considering results of screening non-culture-based tests). Death occurred in 31/122 (25.4%) patients in the micafungin arm and in 28/126 (22.2%) patients in the placebo arm in the safety analysis set (which also included confirmed IC at baseline, registered in 2 and 5 patients in the micafungin and placebo arms, respectively) [50]. In a double-blind RCT conducted in 270 ICU patients with persistent fever despite administration of broad spectrum antibiotics, empirical intravenous fluconazole was compared to placebo with respect to a primary efficacy endpoint of success in all the following: (i) resolution of fever; (ii) absence of invasive fungal infection (iii) no discontinuation owing to toxicity; and (iv) no need for other systemic antifungal medications [51]. Of note, all patients had a high (> 16) APACHE II score and central venous catheters. Overall, achievement of the primary endpoint was registered in 36% (44/122) and 38% (48/127) of patients in fluconazole and placebo arms, respectively (relative risk 0.95, with 95% CI from 0.69 to 1.32) [51].

A summary of results from RCTs on the efficacy of empirical antifungal therapy in ICU patients with suspected IC is also available in Table 1.

Table 1 Primary results of randomized controlled trials evaluating the use of empirical antifungal therapy in critically ill patients with suspected invasive candidiasis

Diagnostic-Driven Early Antifungal Therapy

Theoretically, rapid non-culture-based tests offer two potential advantages in terms of early antifungal therapy in ICU patients with suspected IC: (i) they improve overall PPV for IC in patients with risk scores suggesting IC; (ii) they usually show a high NPV on their own, thereby potentially allowing safe non-initiation or rapid discontinuation of antifungals independent of risk scores results. Various non-culture-based tests have been evaluated for their diagnostic performance in ICU patients, for example, serum 1,3-β-D-glucan (BDG), T2 magnetic resonance Candida assay (T2Candida), multiplex Candida real-time PCR, the combination of mannan antigen (Mn) and anti-mannan IgG antibodies (Anti-Mn), and Candida albicans germ tube antibody (CAGTA). An advantage of these diagnostic tools is their quick turnaround time compared to classical cultures, although they could better be defined as Bayesian biomarkers as they confer a probability of infection and not a definitive diagnosis [30]. For this latter reason, non-culture-based tests should be utilized and interpreted based on pre-test probability of IC (in turn, arising from the combination of clinical picture and baseline risk according to risk factors) [52, 53•, 54].

The most studied non-culture-based test for the diagnosis of IC in critically ill patients is serum BDG, a nearly pan fungal antigen test which has shown variable sensitivity and specificity (mostly around 70–80% and 55–60%, respectively) [53•, 55]. Serum BDG displays an excellent NPV, while its specificity and PPV have been reported to improve with consecutive sampling and by increasing the positivity cut-off (though with the risks of delaying diagnosis and reducing sensitivity and NPV, respectively) [56]. Another suggested way to increase the PPV of BDG in ICU patients with suspected candidemia is to combine serum BDG with other either specific or nonspecific markers of IC, such as low procalcitonin, although the related evidence is still limited to a few observational studies and deserves further investigation [57, 58]. An improved cost-effectiveness of guiding early antifungal therapy by serum BDG results vs. empirical therapy of critically ill patients with suspected candidemia and high Candida score (at least 3 points) has been proposed by using desirability of outcome ranking (DOOR) analyses, although also in this case evidence is still limited and observational [59]. Furthermore, results of DOOR analyses could be complex to interpret at the bedside, by overall suggesting the probability of a “better choice” rather than classical probability of diagnosis [60]. If further explored and validated in the future, the results of DOOR analyses for the early diagnosis of IC could be considered in conjunction with those of classical analyses, with the aim of presenting a more comprehensive evaluation of the patients’ baseline risk of IC to clinicians. Finally, two other general considerations should be made regarding serum BDG as an early diagnostic tool for IC: (i) sensitivity of serum BDG seems to vary according to Candida species, with the lowest having been reported for C. parapsilosis and C. auris [61,62,63]; (ii) there is increasing interest in combining the use of serum BDG with peritoneal BDG in critically ill patients with suspected IAC, with a recent, prospective multicenter study suggesting high NPV in case of low concentrations of both serum BDG and peritoneal BDG [64•].

Regarding molecular methods (including also the T2Candida assay), the experience in critically ill patients is still limited, but certainly their diagnostic performance in this specific population deserves further dedicated investigation [53•, 65,66,67]. Notably, while they are correctly classified as “non-culture-based tests”, they also enable etiological diagnosis at species level, in contrast to antigen-based tests. This could represent a potential advantage, although high quality evidence from well designed RCTs is needed before widespread adoption in clinical practice. Regarding current evidence from RCTs on the use of a diagnostic-driven approach to IC in critically ill patients, the CandiSep trial was a multicenter RCT comparing clinical outcomes of a serum BDG-driven approach versus a culture-driven approach to antifungal prescribing in ICU patients with sepsis or septic shock and at risk of IC (risk defined as total parenteral nutrition, previous renal replacement therapy, abdominal surgery within the last 7 days, and previous antimicrobial therapy for more than 48 h) [68••]. Overall, 339 patients were enrolled with an eventual IC prevalence of 14.2% (48/339). The primary endpoint was 28-day mortality, which occurred in 33.7% (58/172) and 30.5% (51/167) of patients in the BDG-driven and culture-driven arms, respectively (relative risk 1.10, with 95% confidence interval [CI] from 0.80 to 1.51). The median time to antifungal therapy was 1.1 days and 4.4 days in BDG-driven and culture-driven arms, respectively [68••]. In an exploratory subgroup analysis of the previously discussed INTENSE RCT, critically ill patients with intra-abdominal infection and positive serum BDG were more likely to have a confirmed IC than patients with negative serum BDG (odds ratio [OR] 3.66, with 95% confidence interval from 1.01 to 13.29) [50]. However, this was a secondary, exploratory, observational analysis of the data and patients were not randomized to receive antifungals based on BDG results [50]. In the EMPIRICUS RCT, the observed numerically favorable but non-statistically significant impact of micafungin on invasive fungal infection-free survival at day 28 in patients with positive serum BDG (HR 1.41 with placebo as reference, 95% CI from 0.85 to 2.33) comes from a subgroup analysis, and also deserves further investigation [49].

Finally, non-culture-based tests have also been studied for early antifungal discontinuation due to their high NPV, with RCTs suggesting a favorable effect on safely reducing early antifungal therapy duration [69, 70]. Nonetheless, some caution is needed with regard to serum BDG when extrapolating these results to settings with a high relative frequency of C. parapsilosis and/or C. auris as causative agents of IC, due to possible reduced sensitivity [61,62,63].

Pre-emptive Early Antifungal Therapy

The use of a pre-emptive approach to antifungal therapy for IC in non-neutropenic critically ill patients is currently not supported, considering the low PPV of non-culture-based tests for screening purposes in this population, although it should be highlighted that high certainty evidence in this area is currently lacking. Observational studies have provided mixed results, showing on the one hand a possible reduction in the incidence of IC, and on the other hand a possibly high cost in terms of unnecessary antifungal treatment [71,72,73,74]. Evidence from RCTs is also limited. In an RCT conducted in 64 critically ill patients requiring at least 3 days of ICU stay, pre-emptive antifungal therapy with anidulafungin based on results of twice weekly serum BDG screening was compared to empirical therapy (based on physicians’ judgment in patients blinded to serum BDG screening results) [75]. Overall, 21/45 and 5/17 evaluable patients received antifungal therapy in the pre-emptive therapy and empirical therapy arms, respectively. With the important limitation that the study was not designed or sufficiently powered to assess relevant clinical outcomes, survival at ICU discharge, length of ICU stay, proportion of febrile ICU days, and days on broad spectrum antibiotics were similar in the pre-emptive anidulafungin therapy and empirical therapy arms (85% vs. 81%, median 19 vs. 15 days, 8% vs. 12%, and median 3 vs. 4 days, respectively) [75].

Conclusions

Empirical antifungal therapy is always considered at the bedside of critically ill patients with risk factors for IC, owing to the reported increased mortality associated with delays in the start of appropriate antifungal therapy in patients with IC [8,9,10,11,12, 76]. Despite the presence of many well-designed observational studies evaluating empirical antifungal therapy in ICU patients [77,78,79,80,81,82,83,84], in the present narrative review, we decided to focus on results from RCTs, aiming to summarize those results with the highest level of evidence and to better identify current gray areas still deserving further investigation. In this regard, the first consideration to be made is that, while the direction of effect in RCTs exploring efficacy of empirical therapy in ICU patients with suspected IC was most frequently toward a favorable impact of empirical therapy, no formal demonstration of superiority was observed. This latter result stands in contrast to the increased mortality reported from observational studies in case of delayed antifungal therapy in patients with IC, suggesting, in our opinion, that further research is still necessary to better delineate the precise subgroup of ICU patients with suspected IC that would benefit from early antifungal therapy (either early empirical based on risk scores or diagnostic-driven, or a combination of both). In support of this hypothesis, it should be noted that a wide heterogeneity is present regarding the definition of IC risk in RCTs exploring the efficacy of empirical antifungal therapy in ICU patients with suspected IC, a fact that may confound comparison of findings. Furthermore, the prevalence of confirmed IC was low in the enrolled study populations, which raises the possibility of failing to detect a true effect (type 2 error) due to the predominance of patients without IC enrolled in studies. In our opinion, all of this may suggest that a more tailored identification of patients at risk of IC could both improve the outcomes of critically ill patients with IC who receive early antifungal therapy and promote antifungal stewardship. A possible solution to restrict early antifungal therapy to selected subgroups at risk is by using a diagnostic-driven early antifungal therapy approach. The most explored non-culture-based test against this backdrop is serum beta-D-glucan. Results of currently available RCTs show a trend toward a favorable effect of this approach, but are inadequately powered to draw definitive conclusions. In our opinion, further study is still needed to identify both the subgroup of patients that could benefit the most from a diagnostic-driven approach and the best non-culture-based test (or a combination of tests) to be used for this purpose. For example, we recently explored the impact of a machine learning approach to combine the diagnostic performance for IC of serum BDG and procalcitonin with an explainable machine learning prediction of IC based on other nonspecific laboratory results learned on a training set of thousands of patients [85•]. We found a trend toward improved diagnostic performance for IC, but the findings were not statistically significant. Further steps will be to explore the use of other machine learning algorithms to capture more complex relationships across features, and to explore the addition of automatically extracted clinical variables from electronic medical charts to incorporate them in machine learning-based development of early diagnostic tools based on commonly collected laboratory and clinical variables in everyday clinical practice [85•].

While waiting for further dedicated research, in clinical practice empirical antifungal therapy could be considered in critically ill patients with consistent signs and symptoms and a probability of IC > 20–25% based on risk models, or in critically ill patients with septic shock and multiorgan failure, according to expert recommendations from ESICM/ESCMID [56]. Finally, an important note to be made is that part of the heterogeneity in current research on this topic may be related to the lack of a standardized definition of IC (especially non-candidemic) in non-neutropenic critically ill patients, hampering comparison of research findings. To this end, an effort to improve comparability of research studies on IC in ICU, including RCT evaluating the impact and role of any kind of early antifungal therapy, is currently ongoing through the FUNDICU project [53•, 86].