Combined use of serum (1,3)-β-d-glucan and procalcitonin for the early differential diagnosis between candidaemia and bacteraemia in intensive care units

Differentiating candidaemia from bacteraemia is a tough diagnostic challenge for clinicians worldwide, mainly because of their very similar clinical presentation [1‐4]. Although blood cultures remain the mainstay for a definite aetiological diagnosis, their low sensitivity and slow turn-around time have prompted the development of rapid antigen-based diagnostic methods, such as the (1,3)-β-d-glucan (BDG) assay, as useful tools for anticipating the diagnosis of candidaemia and especially for excluding it, strengthening the diagnosis of bacteraemia and favouring the safe discontinuation of useless antifungal treatments [5‐8].

In addition to serum fungal antigens, other less specific markers might help clinicians in the differential diagnosis between candidaemia and bacteraemia. For example, serum procalcitonin (PCT), a well-recognized marker of bacteraemia, has been suggested to be normal or only slightly increased in candidaemia [9‐13]. However, a still unanswered question is how to interpret and balance the results of PCT and BDG when used in combination for the differential diagnosis between the two diseases.

The primary objective of this study was to assess the combined performance of PCT and BDG for the differential diagnosis between candidaemia and bacteraemia in critically ill patients in the intensive care unit (ICU).

This multicentre retrospective study was conducted in three adult ICU wards of two large Italian tertiary care centres; two in Ospedale Policlinico San Martino, a 1200-bed teaching hospital in Genoa, and one in Policlinico Agostino Gemelli, a 1400-bed teaching hospital in Rome. In both hospitals, BDG testing is routinely used for diagnostic purposes in patients with signs and symptoms of infection and considered at risk of candidaemia according to clinical judgement.

From June 2014 to December 2015, all critically ill patients with candidaemia or bacteraemia fulfilling the following criteria were included in the study: 1) 18 years of age or older; 2) BDG testing performed within 48 h before or after the time of the index culture; and 3) serum PCT measured within 24 h before or after the time of the index culture. The different window of time for BDG and PCT was arbitrarily selected in view of the different kinetics in the blood of the two markers with respect to the infectious syndrome [14, 15]. The time of the index culture was defined as the day on which the first blood culture, which later resulted positive for Candida spp. or bacteria, was drawn. Patients with mixed infections (defined as presence of positive cultures for Candida spp. and bacteria less than 48 h apart) were excluded from the study.

During the study period, 166 patients were included in the study (Fig. 1). Of these, 73 had candidaemia and 93 bacteraemia (44% and 56%, respectively). The complete demographic and clinical characteristics of the patients, as well as their possible association with candidaemia or bacteraemia in univariable analysis, are reported in Table 1. Patients with candidaemia were more likely than those with bacteraemia to have a CVC (82% vs. 65%), have been treated with antibiotics (88% vs. 49%), and have a longer hospital stay before the development of the defining infection (median 21 vs. 14 days). The other baseline characteristics were not significantly different between the two populations.

The distribution of BDG and PCT values in patients with bacteraemia and candidaemia is shown in Fig. 2. Most patients were tested for BDG (139/166, 84%) and PCT (143/166, 86%) within 12 h before and 24 h after the positive blood draw. Patients with candidaemia had higher BDG values (median >500 vs. <80 pg/ml, p < 0.001) and lower PCT values (median 0.76 vs. 4.32 ng/ml, p < 0.001) than those with bacteraemia. The optimal discriminatory cut-offs of BDG and PCT for diagnosing candidaemia over bacteraemia in our population were BDG ≥113 pg/ml and PCT <1.93 ng/ml, respectively (Fig. 3). As shown in Table 2, the performance of the optimal BDG cut-off of ≥113 pg/ml (89% sensitivity, 84% specificity, 81% PPV, 91% NPV) was very similar to that of the standard BDG cut-off of ≥80 pg/ml (92% sensitivity, 81% specificity, 79% PPV, 93% NPV), whereas the optimal PCT cut-off of <1.93 ng/ml (84% sensitivity, 69% specificity, 68% PPV, 84% NPV) performed better in terms of sensitivity and specificity for differentiating between the two diseases than a PCT cut-off of <0.5 ng/ml (41% sensitivity, 86% specificity, 70% PPV, 65% NPV). Although with the limits of subgroup analyses and without detecting a statistically significant difference, it is of note that BDG specificity was apparently lower in patients with Candida colonization than in those without (78% vs. 86%, p = 0.514), whereas its sensitivity was similar in those who received antifungals previously and in those who did not (92%% vs 89%%, p = 1). In the multivariable analysis of factors associated with the diagnosis of candidaemia over bacteraemia, BDG ≥113 pg/ml (odds ratio (OR) 31.0, 95% confidence interval (CI) 11.0–87.7; p < 0.001) and PCT <1.93 ng/ml (OR 9.5, 95% CI 3.3–27.1; p < 0.001) were the only two variables independently associated with the diagnosis.

We then assessed the diagnostic performance of BDG and PCT used in combination for the diagnosis of candidaemia over bacteraemia in our cohort. For practical purposes, we tested the combination of the widely used standard BDG cut-off of ≥80 pg/ml (since its performance was very similar to that of the optimal cut-off of ≥113 pg/ml) and the optimal PCT cut-off of <1.93 ng/ml rounded to the nearest ng/ml unit (i.e. <2 ng/ml). Sensitivity, specificity, PPV, NPV, LR+, and LR– of BDG and PCT used in combination for diagnosing candidaemia over bacteraemia are summarized in Table 3. As reported in Table 3, and graphically in Fig. 4, when both markers were indicative of candidaemia (BDG ≥80 pg/ml and PCT <2 ng/ml) they showed a higher PPV for candidaemia (96%) compared to 79% and 66% when BDG and PCT were used separately, respectively. In contrast, when both markers suggested the absence of candidaemia (i.e. when they were both indicative of bacteraemia, BDG <80 pg/ml and PCT ≥2 ng/ml), their NPV for candidaemia was very similar to that of BDG used alone (95% vs. 93%, respectively).

To further characterize the diagnostic performance of BDG and PCT in combination, including the effect of discordant results (i.e. when one marker indicated candidaemia and the other bacteraemia), we used Fagan’s nomograms (Additional file 1: Figure S1). Specifically, Additional file 1 (Figure S1) shows the pre-test and post-test probabilities of candidaemia according to different BDG and PCT results (considered both separately and in combination), at the prevalence of the disease registered in our study (44%). The pre-test and post-test probability of bacteraemia can be calculated as 1 minus the pre-test or post-test probability of candidaemia, respectively. Post-test probabilities of candidaemia and bacteraemia at prevalences different from that registered in our study can be inferred manually on Fagan’s nomograms by intersecting the likelihood ratio (LR) of candidaemia with a straight line starting from the selected pre-test probability of the disease (some examples are provided in Additional file 2, which also propose a possible therapeutic algorithm to be validated in confirmatory studies).

In this retrospective study in a cohort of critically ill patients with candidaemia or bacteraemia, the combined use of BDG and PCT performed better than the two markers used separately for identifying true cases of candidaemia, in the case of concordant results indicating the disease. Conversely, when both markers indicated bacteraemia, their performance was similar to that of BDG used alone, highlighting the high NPV for candidaemia of the BDG test.

Recently, Martínez-Jiménez and colleagues [23] reported on the performance of BDG for the diagnosis of candidaemia in a cohort of 81 patients with either candidaemia or bacteraemia, showing that the performance of the standard BDG cut-off of ≥80 pg/ml for diagnosing candidaemia over bacteraemia (84% sensitivity, 92% specificity, 87% PPV, 90% NPV) was consistent with that reported in the literature for the diagnosis of candidaemia per se [5‐8, 24‐29]. Our results are in line with these findings, since the overall performance of the optimal BDG cut-off for the diagnosis of candidaemia in our population (≥113 pg/ml) was very similar to that of the standard cut-off. In contrast, while a recent meta-analysis showed a PCT threshold of 0.5 ng/ml to be optimal for the diagnosis of bacteraemia [9], the same threshold was inadequate for differentiating bacteraemia from candidaemia in our study, and a higher value (1.93 ng/ml) performed far better for this purpose. This result is not surprising since patients with candidaemia might have slightly increased PCT values [11‐13]. In this regard, the median PCT concentration of 0.73 ng/ml we found in patients with candidaemia is in line with values suggested by other authors for possibly indicating candidaemia [11‐13]. However, it should be noted that in our population the discriminatory performance of PCT was overall inferior to that of BDG (sensitivity 84% and specificity 67% for PCT <2 ng/ml vs. sensitivity 92% and specificity 81% for BDG ≥80 pg/ml). For this reason, in our opinion, PCT should be used cautiously as a single biomarker for diagnosing or ruling out candidaemia when and where BDG is not available.

Our following step was to assess the performance of BDG and PCT used in combination for differentiating between the two diseases. In this regard, we think three possibly useful observations stemming from our study are: 1) in comparison with a positive BDG value considered alone, the presence of both BDG and PCT indicative of candidaemia might strengthen the clinician’s confidence in having identified a true episode of the disease, by increasing specificity and PPV from 81% and 79% to 98% and 96%, respectively; 2) the presence of both BDG and PCT indicative of bacteraemia does not particularly add to the already good ability of BDG alone in ruling out candidaemia, since sensitivity and NPV increased only from 92% and 93% to 96% and 95%, respectively; and 3) in case of discordant BDG and PCT values, the probabilities of candidaemia and bacteraemia might be not particularly altered by the results of the two tests. Given that, in our opinion, the choice on whether discontinuing antifungals in presence of a negative BDG plus low PCT values should be taken very cautiously, at least pending cultures results. Indeed, from a speculative standpoint, we might hypothesise that low PCT values could help unmask some of the rare false-negative BDG results, thus preventing from unsafely discontinuing antifungals. For example, negative BDG results could be found in some cases of C. parapsilosis candidaemia, as we observed in a previous single-centre experience also including non-ICU patients and a higher number of C. parapsilosis isolates [6], albeit this data has to be confirmed in further, prospective studies. However, it remains clear that no definite therapeutic conclusions can be drawn from the present preliminary two-centre retrospective experience, and validation is warranted.

The main drawback of our study is its retrospective nature. Among limitations of the retrospective design is that blood samples for BDG and PCT testing were not collected at standardised points in time. Consequently, some measured values might be different from those present at the time of the index culture, although this bias is reduced by the fact that we mainly included patients who underwent BDG and PCT testing very close to the positive blood draw. Our aim was indeed to include only patients at risk of candidaemia who underwent BDG testing in the presence of signs and symptoms of the disease and pending cultures results, therefore reflecting the diagnostic use of the marker in real life. However, it should also be considered that we could not always retrospectively explore why clinicians suspected the disease. Consequently, we cannot be sure that the high pre-test probability of candidaemia observed in our cohort (44%) reflected only an appropriate use of BDG or was also dependent on a selection bias, and thus unrepresentative of the true prevalence of candidaemia in patients at risk. Furthermore, although it is overall reasonable to approximate the pre-test probability of a certain disease to its local prevalence in selected populations, the former is inherently individual and might increase or decrease because of individual conditions and factors [30]. Sensitivity and specificity of a marker (and thus the LR of a certain disease) might also not always be fixed irrespective of prevalence, a fact that should be considered when extrapolating our results to other prevalences of candidaemia on Fagan’s nomograms [31]. Another limitation is that we did not include mixed infections (candidaemia plus bacteraemia). Although their occurrence was uncommon in our cohort, and thus unlikely to influence the overall diagnostic performance of the combination, this might not be the rule in other hospitals and settings [32]. The low rate of mixed infections we detected might also reflect the short window of time we used for defining an infection as concomitant for diagnostic purposes. Finally, an important aspect should be taken into account when interpreting our results, that is they apply only to critically ill patients at risk of candidaemia, and cannot be extrapolated to critically ill patients without risk factors for fungal infections. However, in our opinion restricting the analysis to patients at risk should be ultimately viewed more as a strength than a limitation.

The combined use of PCT and BDG could be helpful in the diagnostic workflow for critically ill patients with suspected candidaemia. Further studies are needed to understand whether this might significantly impact current therapeutic algorithms, as well as be a cost-effective strategy.