Early diagnosis of candidemia in intensive care unit patients with sepsis: a prospective comparison of (1→3)-β-D-glucan assay, Candida score, and colonization index

Invasive candidiasis (IC) in non-neutropenic patients admitted to the intensive care unit (ICU) is a serious problem [ 1– 3] and, of particular concern, associated with high mortality [ 4], especially when prompt and adequate antifungal treatment is not administered [ 5– 8]. Furthermore, outcome is dependent on the causative Candida species and management of the primary focus of infection [ 9, 10], and continues to be suboptimal [ 11].

Early identification of ICU patients with signs of sepsis at high risk of IC is challenging due to the complexity of the patients' underlying conditions and low yield of fungal cultures [ 12]. However, the timely recognition of IC is essential to driving clinical decision processes and dictating specific therapeutic strategies [ 13].

Measurement of the serum (1→3)-β-D-glucan (BG) is a non-invasive testing for circulating fungal cell wall components [ 14], that allows the systematic screening and prompt identification of fungal infections (with the exception of cryptococcosis and zygomycosis) [ 15]. This test is considered an aid for the diagnosis of fungemia and deep-seated mycoses including IC [ 16– 18], and it appears to be useful for patients with hematological malignancies [ 16, 17]. BG rises before infection becomes clinically apparent [ 16, 17], but the high false-positive rates [ 19, 20] make it necessary to refine its utility as a tool for the early diagnosis of invasive fungal infections (IFIs) [ 15].

At present, only clinical prediction rules of the risk of IC [ 21– 23] or simpler scoring systems that differentiate patients with Candida colonization from those with ongoing but still occult Candida infection [ 24, 25] are available. Their discriminative role varies; for example, a Candida score (CS) ≥3 was shown to select non-neutropenic critically ill patients at risk for IC more accurately than the Candida colonization index [ 25]. Although based on preliminary data, León et al. [ 26] underscored the clinical usefulness of combining CS and serum levels of BG to discriminate between colonization and proven Candida infection.

Of 377 patients with clinical signs of sepsis admitted to the ICU during the study period, 95 who fulfilled the inclusion criteria above specified were enrolled as participants (Figure 1). Characteristics of the 95 subjects (16 case patients in the IFI group and 79 patients in the group without evidence of IFI) are shown in Table 1. The majority of patients (74%) had at least one predisposing defined host (that is, use of corticosteroids or other immunosuppressants) and/or risk factors. Nonetheless, no patients with probable/possible IFI were identified on the basis of modified EORTC/MSG criteria.

The overall prevalence of IFI was 16.8%. Proven cases included 2 pulmonary mold infections (1 with Aspergillus fumigatus and 1 with Fusarium solani) and 14 Candida infections, of which 13 candidemia (10 with C. albicans, 1 with C. glabrata, 1 with C. parapsilosis, and 1 with C. tropicalis) and 1 mediastinitis (with C. albicans). Eighteen patients had documented bacterial bloodstream infections (1 with and 17 without concomitant IFI).

Thirteen of 95 patients (only 1 of the IFI group) were receiving empirical antifungal therapy (7 with echinocandin and 6 with fluconazole) two to four days before testing.

Using a single sample of patients obtained at the onset of the septic syndrome in our evaluation, which included 16 subjects with IFI and 79 with no IFI, a BG level ≥80 pg/mL had sensitivity and specificity of 93.7% and 93.6% for confirmed IFI, respectively. With respect to prior studies [ 15, 34], overall BG assay sensitivity was higher in our cohort, likely because of lack of subjects in the EORTC/MSG categories with inherent diagnostic uncertainty (possible or probable IFI) and/or in whom no active IFI occurred at the time of sampling. Accordingly, BG sensitivity for patients with proven IC was 92.9%, whereas all but five episodes of non-fungal infections had BG concentrations higher than the cut-off value (specificity 91.4%).

In a multicenter evaluation of the Fungitell kit conducted in the United States by Ostrosky-Zeichner et al. [ 35], the sensitivity of the assay for patients with proven IC ( n = 107; cut-off, 80 pg/mL) was 77.6% based on a single sample obtained from the patient within 72 hrs after entry into the study. A single-sample strategy was also adopted in studies that, similarly to ours, involved patients who had or developed systemic Candida infections (for review, see Karageorgopoulos et al. [ 36] and the references therein), in which BG assay sensitivity for proven or probable IC ranged from 85% [ 19] to 92% [ 37]. Although obtaining multiple samples increased specificity and PPV of the BG assay for subjects with acute myeloid leukemia or myelodysplastic syndrome [ 16], this may be not the case for ICU patients, for whom it could be important to explore the performance of the assay with a single-sample protocol. This protocol was believed to be potentially closer to the way in which the assay could be used in ICU clinical settings [ 35].

Our BG sampling strategy involved the use of a blood specimen drawn from an arterial line. In the ICU setting, arterial catheters allow continuous real-time blood pressure monitoring and reduce the need for repeated punctures for blood gas analysis. Therefore, it was reasonable to assess whether this blood source could be exploited for microbiological tests, instead of a central venous catheter (nine of our patients did not harbor this device). This was, also, in view of a recent study showing no significant statistical difference in rates of contamination between venous and arterial sites of blood draw [ 38], at least for patients not particularly vulnerable to fungal colonization/infection, such as burn patients. To date, this is the first evaluation enabling use of the Fungitell kit with sera retrieved from arterial blood specimens, and, in our opinion, information that measurements obtained from either arterial or venous blood samples are equivalent. It could represent a new and very useful practical aspect.

The use of certain intravenous antimicrobials (for example, amoxicillin-clavulanic acid) or gauze packing of serosal surfaces may be a source of false-positive BG measurements [ 39]. We detected a high level of BG in five patients without evidence of fungal infection, one of whom was undergoing extensive intra-abdominal gauze exposure following surgery and four were receiving antimicrobial treatment with β-lactams. However, considering that 4 of the patients with true-negative BG results were exposed to surgical gauze and 36 were treated with β-lactam antibiotics, it is unlikely that a casual relationship between the presence of BG in the blood specimens and medical interventions exists in our patients. Despite the wide use of intravenous albumin or immunoglobulins and/or hemodialysis in ICU patients, such medical treatments did not lead to a false-positive BG elevation in our patients. Some studies also reported a high number of false positives in patients with bacteremia, especially that caused by Gram positive organisms [ 40, 41]. In one study comparing patients with blood cultures positive for yeast (proven or probable IFI) versus bacteremic patients, this appeared to reduce the specificity of the test and to place its PPV at 52%, but given the nature of the samples obtained the authors admitted that BG contamination may have occurred [ 40]. In another study analyzing a single specimen in 46 subjects from ICUs, Digby et al. [ 41] observed that serum BG levels were elevated on average in all patients with either fungal or bacterial infection, even though hemodialysis as a source of false-positivity may be a factor in these observations. By contrast, no cross-reactivity of the BG assay in patients experiencing bacteremias was found in our study and in the prospective validation assays published to date [ 15– 17, 34].

Multisite colonization is a predictor for candidemia in the ICU, especially in combination with other risk factors reflecting severe disease [ 4, 42]. Thus, IC is highly improbable if a Candida- colonized critically ill patient has a CS < 3 [ 26]. Colonization by fungi was seen to not raise the concentration of BG [ 16, 34], while BG may reach the circulation through damaged intestinal or respiratory mucosa and causes positive test results. In this case, BG could be the marker of the gut barrier invasion prior to infection. Of interest, studying a patient population different from ours but with similar risk profiles, Senn et al. [ 17] showed that all of the false-positive BG assay results (4%) were observed in patients with gastrointestinal fungal colonization and/or mucositis who received empirical therapy, suggesting an occult IC at an early stage. Few studies have performed direct comparisons of the BG assay, CS, and colonization index for identifying patients at risk of development of IC [ 26]. In our study, we found that 10 of 13 BG positive patients with proven IC (true positive) had a CS ≥3 and 7 a colonization index ≥0.5, and these findings, on the one hand, reinforce the discriminatory power of BG testing between patients with true IC and those without IC and, on the other hand, suggest its use to make diagnosis more quickly. However, the not too high concordance observed warrants additional investigations whether the three tests can be used as complements to each other, although an analysis of diagnostic performance combining BG assay and CS showed that combined use of the two tests enhanced the sensitivity and NPV for the diagnosis of IC.

Aside from the low frequency of cases of proven IC, the experimental design of our study, giving an estimation of the prevalence of IFI in the target population, allowed an assessment of PPV and NPV by use of BG testing. With a prevalence of fungal infection of 16.8%, the PPV for the BG test was 72.2%, in spite of a very high NPV (approximately 99%). Therefore, a strong diagnostic benefit of BG testing lies in excluding IFI (and IC), according to that reported earlier [ 20, 35, 43], while an unresolved question about the BG testing remains the knowledge of factors that could increase BG levels for reasons other than IFI. Nevertheless, the small number of subjects studied, leading to broad CI intervals, should lower the strength of our observations. Additionally, the pre-test probability of IC in our study population was high given that 74% of the patients presented with risk factors for IFI, thus possibly limiting the generalizability of our data to other patients. However, a strict but effective patient selection strategy such as that adopted in this study, if combined with a single-test approach, may have a major financial impact by reducing the laboratory costs associated with the use of this expensive biomarker.

We are aware of the potential limits of BG assay for detecting fungal infection that would yet balance key deficiencies in the current culture-based methods, such as poor sensitivity and delayed turnaround time. A similar scenario is offered by newer PCR techniques (for example, SeptiFast; Roche Molecular Systems, Mannheim, Germany), though their application to detect and identify fungal pathogens more rapidly than conventional culture might revolutionize the diagnosis and the management of sepsis [ 44]. However, provided an appropriate laboratory logistics, BG assay results can be obtained within a few hours, and then more rapidly than those of PCR techniques, which remain extremely complex and expensive for a routine microbiology laboratory.

The BG assay, based on arterial blood sampling, appears to be useful as a single-point assay for ICU patients with suspicion of Candida sepsis. In addition, a combination of CS and BG measurement in this setting may improve the diagnostic performance and the efficiency of management. However, this strategy deserves further study to assess its true value in early detecting ICs and thus reducing the increased mortality [ 3] and cost of ICU stay [ 45] caused by Candida infections.