Candidemia due to uncommon Candida species in children: new threat and impacts on outcomes

Many uncommon Candida spp. (species other than C. albicans, C. parapsilosis, C. glabrata, C. tropicalis, and C. krusei) have been shown to emerge in tertiary care facilities. We aimed to investigate these uncommon candidemia in children. Forty-six cases of candidemia caused by uncommon Candida spp. were identified during 2003–2015 from a medical center in Taiwan. The most common specie was C. guilliermondii (31.2%), followed by C. lusitaniae (18.8%) and C. metapsilosis (18.8%). These cases were analyzed and compared with 148 episodes of C. albicans candidemia. The incidence density of uncommon Candida spp. candidemia and the proportion to all candidemia episodes increased substantively during the study period. Prior exposure to azoles was uncommon in the 30 days prior to infection, but fluconazole resistant strains were significantly more common (n = 19, 41.3%). The increased incidence density of uncommon Candida spp. candidemia was associated with increasing use of antifungal agents. No differences in demographics, underlying comorbidities, risk factors, clinical features, dissemination, and 30-day mortality were found between uncommon Candida spp. and C. albicans candidemia. Patients with uncommon Candida spp. candidemia were more likely to require modifications in antifungal treatment and receive echinocandin drugs (43.5% vs 21.6%, p = 0.007). Candidemia caused by uncommon Candida spp. had poorer response to antifungal treatment, led to longer duration of candidemia (median 4.0 versus 2.5 days, p = 0.008), and had a higher treatment failure rate (56.5% vs 38.5%, p = 0.040).


Materials and Methods
Study design, collection of isolates and antifungal susceptibility. This study was part of a collaborative, combined retrospective and prospective collected database, laboratory-based, single-center study of invasive yeast infection as previously described 19 . We identified patients younger than 18 years of age with Candida bloodstream infection (BSI) caused by uncommon Candida spp. between January 2003 and December 2015. All Candida isolates were phenotypically identified by using the API 32C AUX yeast identification kit (bioMérieux SA, Marcy l'Étoile, France) and chromogenic culture media (CHROMagar, Becton Dickinson and Company, Franklin Lakes, NJ, USA). Beginning in December 2013, we applied Matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF, Bruker Biotype, software version 3.0, USA), ITS1-5.8S-ITS2 rDNA gene sequencing and large-subunit (18S) ribosomal RNA gene D1/D2 domain sequencing to re-confirm all these species. The study was approved by the Institutional Review Board and Human Research Ethics Committee of Chang Gung Memorial Hospital (CGMH), and a waiver of informed consent for anonymous data collection was also approved. All methods in this current study were performed in accordance with the relevant guidelines mentioned in this manuscript.
We also enrolled all cases of candidemia in children caused by C. albicans during the study period for comparisons. We excluded cases of unidentified Candida spp. and selected only the first isolate recovered from the blood if a patient had several cultures positive for the same Candida spp. Antifungal susceptibility of all these Candida isolates to nine antifungal agents was determined by broth microdilution method using a Sensititre YeastOne system (Trek Diagnostic Systems Ltd., East Grinstead, UK) according to the manufacturer's instructions 20,21 . Minimum inhibitory concentration (MIC) was recorded as the highest concentration of antifungal agent resulting in the development of a blue color. The criteria for susceptibility of these Candida isolates to nine antifungal agents were based on MIC breakpoints of Candida spp. recommended by the Clinical & Laboratory Standards Institute (CLSI) guidelines 22 . For uncommon Candida spp., other than C. guilliermondii, clinical breakpoints are undefined; therefore, isolates that showed MICs higher than the epidemiologic cutoff value were considered potentially resistant 23 .
Data collection and definitions. An incident episode of candidemia was defined as the first positive blood culture drawn from a peripheral vein yielding a Candida species, with clinical symptoms and signs compatible with Candida BSI 24,25 . Episodes were considered to be separate if they occurred at least 1 month apart and when at least one negative blood culture was noted between them. The clinical information was accumulated from a review of medical charts and included demographic characteristics, predisposing risk factors within the preceding 30 days from the onset of Candida BSI (defined as the day of first positive blood culture for Candida spp), underlying diseases, and the presence of an intravenous catheter or any other artificial device at the time candidemia appeared. The clinical manifestations at the time of blood culture collection, ICU admission, and the antimicrobial regimens used were also collected.
An episode of candidemia was considered catheter-related only if the same Candida species was cultured from the catheter tip during the episode; the definition was suggested by the guidelines of the Infectious Diseases Society of America 26,27 . Persistent candidemia was defined as repeated positive blood cultures for Candida spp. for more than 3 days after antifungal agents were administered. Candidemia-attributable mortality was defined as patients died within 7 days after onset of candidemia or in the presence of persistent clinical sepsis or persistent candidemia, or those died of candidemia associated complications 28,29 . Breakthrough candidemia was defined as new occurrence of candidemia while the patient was on antifungal agents for more than three days 30,31 . The primary study outcome was clinical treatment failure, which was defined according to the Mycoses Study Group and European Organization for Research and Treatment of Cancer consensus criteria 32 as the following: (1) all-cause mortality between days 3 and 30 after the initial positive blood culture, or (2) persistent fungal BSI for ≥72 hours after the initiation of antifungal therapy. The secondary outcome was all cause in-hospital mortality. Patients' response to antifungal therapy following candidemia was defined according to the consensus criteria of the Mycoses Study Group and European Organization for Research and Treatment of Cancer 33 .
Statistical analysis. The demographic, clinical, outcome variables and the in vitro susceptibility data were summarized using the descriptive statistics. All statistical analyses were performed using IBM SPSS software (version 22.0; IBM SPSS Inc., New York, USA). Categorical variables were compared using the χ 2 or Fisher's exact test, and continuous variables by the Mann-Whitney U test. A P value of 0.05 was considered significant. Poisson regression and the Cochran-Armitage test were used for trend analysis of the annual BSI incidence densities and the proportions of candidemia caused by uncommon Candida spp., respectively. We also compared BSI incidence densities for three time periods; 2003-2006, 2007-2011, and 2012-2015, using Poisson distribution and test-based methods. The correlation between the annual use of antifungals and time was evaluated by using the Spearman correlation. The associations between the incidence densities of uncommon Candida spp. BSI and the annual use of antifungals (defined as daily doses per 1,000 patient-days) were evaluated by using Poisson regression.
We used Cox regression analysis to identify factors that were significantly associated with death. Clinically relevant parameters in the univariate analysis (P < 0.1) were included at multivariate regression analysis. The full model was reduced to a final model by using a stepwise elimination procedure. The proportional hazards assumption was tested graphically and by building time-dependent variables.

Results
We identified 323 episodes of candidemia that occurred in hospitalized children over the 13-year study period. A total of 25 cultures that previously grew unspecified Candida spp. were rechecked by our ITS1-5.  (Table 1). These data were compared with those reported for 148 episodes of candidemia caused by C. albicans in 136 patients.
The overall incidence of uncommon Candida spp. candidemia and their proportion relative to all episodes of candidemia increased significantly during 2003-2015 (incidence density p < 0.001; proportion p < 0.001) (Fig. 1). The overall incidence density of uncommon Candida spp. BSI was 3.61 episodes per 100,000 inpatient days, which increased from 1.48 (2003-2006)  and was significantly associated with the continuous increase in overall antifungal agent use (p = 0.033).
Most chronic comorbid conditions and associated risk factors were comparable between cases of candidemia caused by uncommon Candida spp. and C. albicans groups ( Table 2). Although uncommon Candida spp. had a significantly higher MIC to fluconazole and relatively more commonly presented as breakthrough candidemia (17.4% vs 8.1%, P = 0.094), previous exposure to azoles was comparable between these two groups. The clinical characteristics, therapeutic regimens and treatment responses of uncommon Candida spp. candidemia and C. albicans candidemia are compared in Table 3. The severity of illness, judged by rates of severe sepsis and septic shock, were comparable between cases of candidemia caused by uncommon Candida spp. and those due to C. albicans. However, candidemia caused by uncommon Candida spp. led to significantly higher rates of persistent candidemia compared to C. albicans candidemia (76.1% vs 56.8%, p = 0.024).
Overall, 186 episodes (95.9%) of candidemia were treated with specific antifungal agents. The mean duration between onset of candidemia (time of the first positive blood culture for Candida spp.) and initiation of antifungal agents was 2.3 days (range 0-8 days). In 83 episodes (44.6%) of the 186 episodes, antifungal regimens were modified during the treatment course. Patients with uncommon Candida spp. candidemia had a significantly higher rate of antifungal regimens modification than did those with C. albicans candidemia, and were more often treated with echinocandins (43.5% vs 21.6%, p = 0.007). After modification of antifungal regimens, uncommon Candida spp. led to a significantly longer duration of candidemia (median 3.0 versus 1.0 days, p < 0.001), were slower to respond to antifungal agents, and had significantly higher rates of clinical treatment failure (56.5% versus 38.5%, p = 0.040), although the deaths due to fungemia and final in-hospital mortality were comparable between these two groups.
MICs for uncommon Candida spp. of eight antifungal agents are shown in Table 4

Discussion
Population-based surveillance studies have documented the shift of candidemia from C. albicans to non-albicans species over the past two decades 2,4,5 . Uncommon Candida species, which generally account for less than 10% of all cases of candidemia 17,18 , are emerging among critically ill patients. The reported prevalence of uncommon Candida spp. in children varies widely between 3.2-22%, depending on definitions, geographic region, and patient characteristics [34][35][36][37] . We found that uncommon Candida spp. accounted for 14.4% of all cases of candidemia in children, and our result is in agreement with a recent study that documented the incidence and proportion of uncommon Candida spp. BSIs has risen during the past decade. We found uncommon Candida spp. candidemia were more frequently associated with treatment failure than candidemia caused by C. albicans, as these isolates were more commonly resistant to azoles, which led to poorer response and longer duration of candidemia.
In contrast to non-albicans candidemia in adults, in which prior fluconazole exposure was often concluded as an independent risk factor 18,38,39 , studies in the pediatric populations found no difference between C. albicans and non-albicans candidemia in terms of demographics, underlying disease, risk factors, clinical features and outcomes 32,40 . However, previous studies attempting to investigate the risk factors of non-albicans candidemia in children have focused on the more common Candida spp., such as C. parapsilosis and C. glabrata 32,41,42 . To our knowledge, this is the first study to investigate uncommon Candida spp. candidemia in children. We found similar patients characteristics, risk factors and comparable outcomes between uncommon Candida spp. and C. albicans. These results are in agreement with other reports concluding host characteristics and underlying medical illness are the most powerful predictors of final outcomes 14,[42][43][44] .
The overall incidence of candidemia during the study period remained stable and was not affected by the changes in antifungal treatment policies 45,46 . It is tempting to speculate that an increase in strains resistant to fluconazole and higher MICs values were associated with the changes of in antifungal treatment policies. Published guidelines encourage empiric use of echinocandins in patients with severe illness, history of azole exposure, or neutropenia 47 . This may have influenced the anti-fungal prescribing practices. It is possible that a significant increase in consumption of antifungal agents during the first half of the study period accounted for the emergence of uncommon Candida species, which required longer periods of antifungal treatment and led to a vicious cycle of more uncommon Candida species. Therefore, the incidence density of uncommon Candida spp. BSIs was noted to be significantly higher after 2012.
During the 13-year study period, the approach and antifungal treatment policies at our institute changed in two aspects. In the neonatal intensive care unit, antifungal prophylaxis with fluconazole for extremely low birth weight infants was launched in 2011, and echinocandins became available since 2004. Caspofungin (Cancidas ® , Merck, Sharp & Dohme, Kenilworth, NJ, USA) has been widely used since 2005 and micafungin (Micamine ® , Astellas Pharma, Inc., Tokyo, Japan) became more common beginning in 2009. These changes in antifungal regimens may account for the changing epidemiologic characteristics of candidemia in children since 2011. Several studies have concluded that the increase of certain non-albicans or uncommon Candida species, such as C. glabrata and C. kefyr, are associated with the increasing use of echinocandin drugs 17,42,45 . Another study found significant positive correlation between use of itraconzaole and the increased incidence of C. parapsilosis and C. guilliermondii candidemia 45 . However, no antifungal agent can account for the emergence of uncommon Candida spp. candidemia found in this study. In addition, our cases of uncommon Candida spp. candidemia were less commonly breakthrough candidemia 17 or due to previous treatment with specific antifungal agents.
In several studies, C. guilliermondii has been the most commonly isolated uncommon Candida spp. among pediatric patients 16,24,48 , and C. lusitaniae was common in another international study in children 33 . In other recent reports, more than half of all patients with candidemia caused by uncommon Candida spp. had breakthrough infections or underlying hematological malignancies 15,17 . In our cohort, almost all cases of pediatric candidemia had specific chronic comorbidities and the majority of breakthrough candidemia cases were due to C. parapsilosis and C. glabrata. These differences are a further reflection of the changing epidemiologic characteristics of pediatric candidemia and unique features of uncommon Candida spp. in children. Therefore, we concluded that uncommon Candida spp. distributions and clinical characteristics vary by patient population, geographic region, and antifungal practices 15,17,49 .
Our study had some limitations. First, it was a retrospective study from a single institution with a small number of episodes caused by individual uncommon Candida spp.; therefore, further multicenter, prospective studies, or systemic review with meta-analysis are required to update information applicable to different geographic areas or specific groups at risk for candidemia caused by uncommon Candida species. Second, although we used the MALDI-TOF and DNA sequencing to re-identify all Candida isolates in the past 13 years, some Candida strains of pediatric candidemia more than five years ago were not available and were identified phenotypically at that time. Therefore, it is possible that during 2003-2011, some C. dublinensis and other Candida isolates were mis-identified as C. albicans, and the frequency of uncommon Candida spp was thus underestimated. Although MALDI-TOF has strengths of rapid, sensitive, and economical in terms of both costs and labor involved, it is also limited that the spectral database of MALDI-TOF must contain peptide mass fingerprints of the specific species before it can correctly identify new species 50 . Finally, the uncommon Candida species are a heterogeneous "mixture" of many different organisms. Therefore, they may not share common clinical characteristics and treatment strategies should depend on individual cases.  In conclusion, uncommon Candida spp. causing candidemia are emerging among hospitalized children. We did not find clinical variables that enable us early recognition or prediction of candidemia caused by these pathogens. Although clinical outcomes at day 30 are similar to those caused by C. albicans, uncommon Candida spp. more frequently result in prolonged fungemia and treatment failure. Because uncommon Candida species frequently show fluconazole MICs above the epidemiologic cutoff values, identification of all Candida organisms at the species level by advanced molecular methods is of value in guiding treatment directions.

Conclusion
Uncommon Candida species have now emerged among hospitalized children. These pathogens frequently are not susceptible to fluconazole and had higher rate of treatment failure; echinocandins are the treatment choice.