A prediction rule for early recognition

of patients with candidemia in Internal

Medicine: results from an Italian,
multicentric, case–control study

on behalf of GISA/FADOI Candida

Study Group

Infection
A Journal of Infectious Diseases

ISSN 0300-8126

Infection
DOI 10.1007/s15010-018-1162-0

1 23
Your article is protected by copyright and
all rights are held exclusively by Springer-
Verlag GmbH Germany, part of Springer
Nature. This e-offprint is for personal use only
and shall not be self-archived in electronic
repositories. If you wish to self-archive your
article, please use the accepted manuscript
version for posting on your own website. You
may further deposit the accepted manuscript
version in any repository, provided it is only
made publicly available 12 months after
official publication or later and provided
acknowledgement is given to the original
source of publication and a link is inserted
to the published article on Springer's
website. The link must be accompanied by
the following text: "The final publication is
available at link.springer.com”.

1 23
 Author's personal copy
Infection
https://doi.org/10.1007/s15010-018-1162-0

ORIGINAL PAPER

A prediction rule for early recognition of patients with candidemia

in Internal Medicine: results from an Italian, multicentric, case–control
study
Emanuela Sozio1 · Filippo Pieralli2 · Anna Maria Azzini3 · Giancarlo Tintori4 · Federica Demma5 · Gianluca Furneri5 ·
Francesco Sbrana6 · Giacomo Bertolino7 · Simona Fortunato8 · Simone Meini9 · Damiano Bragantini3 ·
Alessandro Morettini10 · Carlo Nozzoli10 · Francesco Menichetti8 · Ercole Concia3 · Carlo Tascini11 · on behalf of GISA/
FADOI Candida Study Group

Received: 17 March 2018 / Accepted: 5 June 2018

© Springer-Verlag GmbH Germany, part of Springer Nature 2018

Abstract
Purpose  Increasing prevalence of candidemia in Internal Medicine wards (IMWs) has been reported in recent years, but risk
factors for candida bloodstream infection in patients admitted to IMW may differ from those known in other settings. The
aim of this study was to identify risk factors and define a prediction rule for the early recognition of the risk of candidemia
in IMW inpatients.
Methods  This was a multicentric, retrospective, observational case–control study on non-neutropenic patients with can-
didemia admitted to IMWs of four large Italian Hospitals. Each eligible patient with candidemia (case) was matched to a
control with bacteremia. Stepwise logistic regression analyses were performed.
Results  Overall, 300 patients (150 cases and 150 controls) were enrolled. The following factors were associated with an
increased risk of candidemia and weighted to build a score: total parenteral nutrition (OR 2.45, p = 0.008; 1 point); central
venous catheter (OR 2.19, p = 0.031; 1 point); peripherally inserted central catheter (OR 5.63, p < 0.0001; 3 points), antibiotic
treatment prior (OR 2.06; p = 0.059; 1 point) and during hospitalization (OR2.38, p = 0.033; 1 point); neurological disability
(OR 2.25, p = 0.01; 1 point); and previous hospitalization within 3 months (OR 1.56, p = 0.163; 1 point). At ROC curve
analysis, a final score ≥ 4 showed 84% sensitivity, 76% specificity, and 80% accuracy in predicting the risk of candidemia.
Conclusions  The proposed scoring system showed to be a simple and highly performing tool in distinguishing bloodstream
infections due to Candida and bacteria in patients admitted to IMW. The proposed rule might help to reduce delay in empiri-
cal treatment and improve appropriateness in antifungal prescription in septic patients.

Keywords  Candidemia · Internal medicine wards · Prediction rule · Risk factors

Introduction currently between the fourth and the sixth most common
nosocomial bloodstream isolates found in studies from the
Candidemia is an important cause of bloodstream infec- United States and Europe [1–9], representing around the 9%
tions (BSIs), leading to significant mortality and morbidity of all nosocomial BSIs [1]. In Italy, its incidence is estimated
in health-care settings. The global incidence of candidemia between 0.8 and 2.53 episodes per 1000 hospital admission
increased fivefold in the past 15 years, and Candida spp. are [10, 11]. That variability may reflect the different profile
of risk factors for candidemia, including demographics,
patients’ case mix, and different use of invasive procedures
Electronic supplementary material  The online version of this [12].
article (https​://doi.org/10.1007/s1501​0-018-1162-0) contains
A retrospective cohort study conducted in Intensive Care
supplementary material, which is available to authorized users.
Unit (ICU) among patients with candidemia shown that
* Emanuela Sozio candidemia acquired after 48-h ICU stay had significantly
emanuela.sozio@gmail.com more frequent undergone previous surgery and organ failures
Extended author information available on the last page of the article with cardiovascular, renal, central nervous, and coagulation

13
Vol.:(0123456789)
 Author's personal copy
E. Sozio et al.

systems than candidemia acquired before ICU admission or acute pancreatitis, broad-spectrum antibiotic and antifungal
during the first 48-h [13]. treatment in the previous 30 days, concomitant antibiotic
In recent years, an increasing number of Candida BSIs therapy, steroids or other immunosuppressant treatment,
(up to 59% of all nosocomial candidemia) has been observed major surgical interventions in previous 3 months, dialysis
in patients admitted to Internal Medical Wards (IMWs), and C. difficile infection (CDI), and data about therapy and
where patients are usually old, with multiple comorbidities, laboratory were recorded. Candida colonization defined both
and often have multiple risk factors for candidemia [10, 11, by the Candida score and the Ostrosky-Zeichner prediction
14–18]. Prediction rules for invasive candidiasis, such as the rule was collected as well [19, 20].
Candida score and Ostrosky-Zeichner score, have been for- In cases and controls, all risk factors were assessed at
mulated to predict the risk in surgical and medical patients the onset of candidemia/bacteremia (date of collection of
admitted to ICU [19–24]. However, no risk score has been the blood sample subsequently resulted positive). All col-
expressly carried out to identify the patients admitted to lected data have been de-identified to preserve participant’s
IMWs at risk of candidemia. anonymity and confidentiality.
Aim of this study was to identify risk factors for candi-
demia and develop a prediction rule to assist clinicians to Definitions
the early recognition of the risk of Candida bloodstream
infection in septic patients admitted in IMWs. A case of candidemia was defined as a patient with at least
one blood culture yielding Candida species.
A case of bacteremia was defined as at least one blood
Materials and methods culture yielding a Gram-negative or Gram-positive bacte-
ria; in case of coagulase-negative staphylococci, at least two
Study design and population consecutive blood culture sets yielding the same strains were
required. Blood cultures positive for more than one species
In this multicenter, retrospective, case–control study were were excluded.
included all consecutive adult patients with candidemia We included patients with fever and/or other clinical signs
admitted in IMWs of Pisa, Firenze, and Verona Hospitals of infection (basing on SIRS criteria). Clinical severity at the
(Italy), between February 2012 and August 2015. Only cases diagnosis of candidemia/bacteremia was assessed according
occurred in IMWs were included. to sepsis grading purposed by the Surviving Sepsis Cam-
Cases were retrospectively identified through the micro- paign [25].
biology laboratory database; in the case of more than one
episode of candidemia in the same patient, only the first Microbiology laboratory methods
episode of candidemia was considered for the study.
For each case, one control in the same IMW, reporting Blood cultures were processed using the automated blood
bacteremia and matched for age (± 5 years), sex, date of hos- culture system BacT/Alert 3D (BioMerieux Inc., Marcy-
pital admission, and duration of hospitalization (± 20 days) l’Etoile, France). Positive cultures were sub-cultured
at time of first positive blood culture, was selected in a 1:1 and identified to the species level by VitekMS or Vitek2
ratio. To ensure comparable periods of risk exposure in both (BioMerieux; Marcy-l’Étoile, France). Susceptibility testing
groups, each control had a length of hospitalization similar was performed using Vitek2 system and interpreted accord-
to the time at risk of case (defined as the number of days ing to EUCAST criteria effective during the study period.
from hospital admission to candidemia occurrence). Neu-
tropenic patients (absolute neutrophil count of < 1000 cells/ Statistical analysis
mm3) were excluded. This study was exempt from institu-
tional review board oversight because of its retrospective Standard descriptive statistics were used to analyze patients’
nature and the anonymity of pooled data. characteristics at baseline. All continuous variables were
expressed as mean and standard deviation (SD). Categori-
Data cal data were presented as percentages. For the unadjusted
analysis, univariate logistic regressions were run. Variables
Cases and controls clinical records were used to collect associated with a p value < 0.20 [23, 26, 27] were consid-
demographic and clinical data; comorbidities were assessed ered significant and included in multivariate analysis. For
by the Charlson comorbidity index. Information about inva- the adjusted analysis, multivariate logistic models were real-
sive devices (central venous catheter—CVC; peripherally ized to identify independent predictors of candidemia. The
inserted central catheter—PICC; bladder catheter—BC; model with the highest performance was selected. Sensi-
nasogastric tube—NGT), total parenteral nutrition (TPN), tivity analysis was performed to test the robustness of the

13
 Author's personal copy
A prediction rule for early recognition of patients with candidemia in Internal Medicine: results…

selected model. A candidemia scoring system was created by tropicalis (7.3%), C. glabrata (6.7%), C. krusei (2%), C.
rounding the odds ratios (ORs) of the chosen model [28, 29]. famata (2%), and C. lusitaniae (1.3%).
ROC curve was used to determine the optimal cutoff score to Among controls, Gram-negative strains were 51% and the
discriminate between the two states “high risk of candida” other 49% were Gram-positive.
and “high risk of any bacterial infection”. The median age of the whole population was 76 years;
More details on statistical analysis are shown in Supple- sex distribution was balanced, with 48.7% of women.
mentary Materials. Diagnostic criteria for severe sepsis/septic shock were
present in 48.3% of patient population.
In Table 1 are described demographic and clinical fea-
Results tures of the enrolled population.
In general, candidemia was a delayed event with respect
A total of 300 patients with candidemia or bacteremia (150 to bacteremia, since it occurred after a mean of 9 days from
cases and 150 controls) were included in the study. admission (9 days ± 10.96, CI 7.27–10.80) versus a mean
Candida albicans was the most commonly documented of 3.7 days for bacteremia (3.72 days ± 8.33, CI 2.37–5.06)
species (58.7%), followed by C. parapsilosis (22%), C. (p < 0.001).

Table 1  Comparison of patients with candidemia (cases) and patients with bacteremia (controls) admitted in internal medical wards
Variables Global population (N = 300) Patients with candidemia Patients with bac-
(cases) (N = 150) teremia (controls)
(N = 150)

Age [mean; median (SD)] 73.5; 76.8 (14.8) 74.3; 77.7 (14.6) 72.6; 76.1 (15.0)
Sex (female) 146 (48.7%) 73 (48.7%) 73 (48.7%)
Diagnostic criteria for severe sepsis/septic shock 145 (48.3%) 86 (57.3%) 59 (39.3%)
Fever 260 (86.7%) 123 (82.0%) 137 (91.3%)
Charlson score [mean; median (SD)] 5.3; 5 (3.5) 6.1; 6 (3.4) 4.5; 4 (3.4)
Ostrosky-Zeichner score 95 (31.7%) 65 (43.3%) 30 (20.0%)
Candida score [mean; median (SD)] 1.8; 2 (1.4) 2.3; 2 (1.3) 1.3; 1 (1.2)
Days to blood culture [mean; median (SD)] 6.4; 2.0 (10.1) 9.0; 5.5 (11.0) 3.7; 0.0 (8.3)
Hospitalizations in the previous 3 months 178 (59.3%) 108 (72.0%) 70 (46.7%)
Previous antibiotic treatment 177 (59.0%) 117 (78.0%) 60 (40.0%)
Previous C. difficile infection 19 (6.3%) 15 (10.0%) 4 (2.7%)
Previous antifungal treatment 20 (6.7%) 11 (7.3%) 9 (6.0%)
Immunosuppressants 22 (7.3%) 7 (4.7%) 15 (10.0%)
Steroids during hospitalization 121 (40.33%) 75 (50.0%) 46 (30.7%)
Antibiotics during hospitalization 187 (62.3%) 122 (81.3%) 65 (43.3%)
TPN 107 (35.7%) 83 (55.3%) 24 (16.0%)
NGT 64 (21.3%) 45 (30.0%) 19 (12.7%)
PICC 87 (29%) 70 (46.7%) 17 (11.3%)
CVC 92 (30.7%) 49 (32.7%) 43 (28.7%)
BC 190 (63.6%) 116 (77.3%) 74 (49.3%)
Chronic kidney disease 68 (22.7%) 25 (16.7%) 43 (28.7%)
Diabetes mellitus 95 (31.7%) 42 (28.0%) 53 (35.3%)
Liver disease 31 (10.3%) 18 (12.0%) 13(8.7%)
Cancer 78 (26.0%) 42 (28.0%) 36 (24.0%)
Dementia 59 (19.7%) 43 (28.7%) 16 (10.7%)
Cerebrovascular disease 86 (28.7%) 57 (38.0%) 29 (19.3%)
Severe functional impairment as Hemiplegia 37 (12.3%) 26 (17.3%) 11 (7.3%)
Ischemic heart disease 50 (16.7%) 17 (11.3%) 33 (22.0%)
Peripheral vascular disease 67 (22.3%) 41 (27.3%) 26 (17.3%)
Congestive heart failure 44 (14.7%) 18 (12.0%) 26 (17.3%)

TPN total parenteral nutrition, PICC peripherally inserted central catheter, CVC central venous catheter, BC bladder catheter, NGT nasogastric
tube

13
 Author's personal copy
E. Sozio et al.

By a first comparison between cases and controls, fac- patients with PICC, and a moderate to strong risk in patients
tors significantly associated with a moderate to strong receiving TPN, previously hospitalized, in those with previ-
risk of candidemia were: TPN (OR = 6.50; p < 0.001); ous and/or concomitant antibiotic therapy, in the presence of
presence of an intravascular device such as PICC or CVC a CVC or affected by neurological disability.
(OR = 5.75; p < 0.001), with a very strong association with Table 3 shows the OR of all the variables of the better
PICC (OR = 6.84; p < 0.001) respect to CVC (OR = 1.20; performing statistical model (see more details on Supple-
p = 0.4530); previous antibiotic therapy (OR = 5,31; mentary Materials).
p < 0.001) and antibiotic treatment during hospitalization Using the best performing statistical model (see more
(OR = 5.69; p < 0.001); C. difficile infection in the previous details on Supplementary Materials), a new prediction rule
30 days (OR = 4.05; p = 0.0150); BC (OR = 3.5; p < 0.001); for candidemia for septic patients hospitalized in IMWs
presence of neurological disability such as cerebrovascular was built up. Table 4 summarizes this score and the points
disease, dementia, hemiplegia (OR = 3.0; p < 0.001); hospi- assigned to each variable.
talization in the previous 3 months (OR = 2.94; p = 0.0150), Using ROC curve analysis, the threshold to define a high
and presence of a NGT (OR = 2.95; p < 0.001). risk for candidemia was a score greater or equal than 4. A
Table 2 describes univariate (unadjusted) analysis of risk patient with a final score ≥ 4 was at high risk to be affected
factors for candidemia and bacteremia in our population. by candidemia; a patient reporting a final score < 4 was at
The OR of all the variables resulted significant at the 0.20 high risk to be affected by bacteremia.
level were rounded and used as points assigned to each vari- Table 5 summarizes performances of the score.
able (see more details on Supplementary Materials). From A score ≥ 4 correctly identified 126 out of 150 candidem-
these data and with respect to the better performing statisti- ias (84% true positive) and misidentified as fungal infections
cal model, we deduced a very strong risk of candidemia in 36/150 bacteremias (24% false positive). On the other hand,

Table 2  Univariate (unadjusted) Variables Candidemia Bacteremia OR 95% CI p value

analysis of risk factors for (N = 150), % (N = 150), %
candidemia and bacteremia
in Internal Medicine Wards TPN 55.3 16.0 6.50 3.78–11.19 < 0.0001
Intravascular devices (PICC and CVC) 79.3 40.0 5.75 3.45–9.61 < 0.0001
PICC 46.7 11.3 6.84 3.76–12.45 < 0.0001
Previous antibiotic therapy 78.0 40.0 5.31 3.20–8.82 < 0.0001
Antibiotics during hospitalization 81.3 43.3 5.69 3.38–9.60 < 0.0001
BC 77.3 49.3 3.50 2.13–5.77 < 0.0001
Neurological ­Disabilitya 54.7 28.7 3.00 1.86–4.84 < 0.0001
Hospitalization in the previous 3 months 72.0 46.7 2.94 1.82–4.75 < 0.0001
NGT 30.0 12.7 2.95 1.63–5.35 < 0.0001
Steroids during hospitalization 50.0 30.7 2.26 1.41–3.62 0.0001
Chronic kidney disease 16.7 28.7 0.50 0.28–0.87 0.0140
Previous C. difficile infection 10.0 2.7 4.05 1.31–12.52 0.0150
Fever 82.0 91.3 0.43 0.21–0.87 0.0200
Immunosuppressants 4.7 10.0 0.44 0.17–1.11 0.0830
Diabetes mellitus 28.0 35.3 0.71 0.44–1.16 0.1730
Liver disease 12.0 8.7 1.43 0.68–3.05 0.3450
Cancer 28.0 24.0 1.23 0.73–2.06 0.4300
CVC 32.7 28.7% 1.20 0.74–1.97 0.4530
Previous antifungal therapy 7.3 6.0 1.24 0.49–3.08 0.6440
Cardiovascular ­diseasec 40.0 39.3 1.02 0.65–1.63 0.9600
Female sex 48.7 48.7 1.00 0.63–1.58 1.0000
Age ≥ 63.07b 80.0 74.0 1.41 0.82–2.41 0.2180

TPN total parenteral nutrition, PICC peripherally inserted central catheter, CVC central venous catheter,
BC bladder catheter, NGT nasogastric tube
a
 Presence of at least one of the following conditions: cerebrovascular disease, dementia, and hemiplegia
b
 Threshold obtained using ROC curve
c
 Presence of at least one of the following conditions: ischemic heart disease, congestive heart failure, and
peripheral vascular disease

13
 Author's personal copy
A prediction rule for early recognition of patients with candidemia in Internal Medicine: results…

Table 3  Selected model and candidemia prediction rule Thus, candidemia is an important challenge for Intern-
Variable OR p value Rounded ists, who should be aware of it, since the high mortal-
coeffi- ity and morbidity rates prolonged hospitalization and
cient increased costs for the entire health-care system [6, 30].
Despite the increasing incidence of candidemia in
Hospitalizations in the previous 3 months 1.56 0.1630 2
patients admitted in IMW, there is to date scant evidence
Previous antibiotic treatment 2.06 0.0590 2
on the characterization of risk factors on this specific
Antibiotics during hospitalization 2.38 0.0330 2
population.
Immunosuppressant 0.40 0.1030 0
In this study, we describe risk factors independently
CVC 2.19 0.0310 2
associated with an increased risk of candidemia in a pop-
PICC 5.63 0.0000 6
ulation of patients admitted and entirely managed in an
TPN 2.45 0.0080 2
IMW.
Neurological disability 2.25 0.0100 2
At the ultimate analysis, the presence of a central venous
Kidney failure 0.68 0.2780 NS
line, TPN, hospitalization during the previous 3 months,
Fever 0.71 0.4740 NS
previous (within 1 month) or ongoing antibiotic therapy,
Steroids 1.14 0.6740 NS
and severe neurological disability (including dementia and
Clostridium difficile infection 1.35 0.6560 NS
invalidating stroke) represent factors independently associ-
Constant 1.56 0.1630 -
ated with a significant increasing risk for candidemia. All
All the variables assumed value 1 if Yes; 0 otherwise these factors increase the probability of candidemia by a
CVC central venous catheter, PICC peripherally inserted central cath- factor of 2, whereas the presence of a PICC increases that
eter, TPN total parenteral nutrition, NS not significant so not consid- probability by a factor of 5.
ered in the scoring system
Available data about risk profiles for candidemia mainly
derive from ICU and surgical wards, where patients’ charac-
Table 4  Risk score for candidemia in patients hospitalized in IMWs teristics are deeply different from those of patients admitted
Risk factor Points
in IMWs, for the burden of comorbidities, and demographic
and health-care-related factors.
PICC + 3 For these and other reasons, scores designed for ICU or
CVC + 1 surgical patients (i.e., Ostrosky-Zeichner and Leon candida
TPN + 1 score) [19, 20] are realistically unsuitable for the applica-
Neurological ­disabilitya + 1 tion in IMWs, where a large part of candidemias occur, and
Hospitalization in the previous 3 months + 1 our results seem of value adding new insight in a setting for
Previous antibiotic therapy + 1 which data are currently lacking.
Antibiotics during hospitalization + 1 To predict the risk of candidemia in IMWs setting, we
PICC peripherally inserted central catheter, CVC central venous cath-
derived a prediction rule that showed a high NPV (82.6%)
eter, TPN total parenteral nutrition with a favorable PPV (77.8%), a global accuracy of 80%,
a
 Presence of at least one of the following conditions: cerebrovascular and a good performance in terms of sensitivity (84%) and
disease, dementia, and hemiplegia specificity (76%).
This prediction rule would have permitted to correctly
identify 126 out of 150 candidemias (84% of true positives)
it misidentified as bacterial infections 24 out of 150 candi- and to misidentify as fungal infection 36/150 bacteremias
demias (16% false negative). PICC was absent in all of these (24% of false positives). On the contrary, it would have misi-
false negative patients. dentified as bacterial infections, 24/150 candidemias (16%
of false negatives). We consider this latter group of utmost
importance, because even in the presence of a documented
fungemia they would have been probably misinterpreted and
Discussion not treated with antifungal agents, according to this score, at
least until blood culture turned positive.
The incidence of candidemia is growing in patients admitted A score greater or equal than 4 would be able to discrimi-
to IMWs because of demographic changes, with growing nate patients that would need an early empiric antifungal
numbers of frail elderly patients affected by multiple comor- therapy, but we suggest that all septic patients should any-
bidities, and the increasing use of invasive procedures and way start an empirical antibiotic therapy, since even with
complex surgical techniques that subsequently increase the a score ≥ 4 about 25% of patients with BSI due to bacteria
risk of health-care-associated infections [1–12]. could be misidentified as fungal infection.

13
 Author's personal copy
E. Sozio et al.

Table 5  Performance of the proposed score

Prediction rule for a score ≥ 4
Number of observed Number of predicted
Candidemia Bacteremia

Candida BSIs (N = 150) 126 24

Bacterial BSIs (N = 150) 36 114
Performance indicators

PPV 77.8%
NPV 82.6%
Sensitivity 84.0%
Specificity 76.0%
Accuracy 80.0%

PPV positive predictive value, NPV negative predictive value

Then, our score could be useful as a decision support frail population at a higher risk of death and complications
for the Internists in starting an empiric antifungal therapy independently of severity and origin of infection.
in addition to antibiotic therapy in front of a patient with In general, multiple factors account for the increased
sepsis. susceptibility of older patients to infections, including
If the score is lower than 4, the antifungal therapy could immune senescence, institutionalization, and concomitant
be delayed, since the high NPV (82.6%). chronic comorbidities [12]. All these factors expose subjects
Of note, PICC was absent in all the 24 falsely negative to endogenous or exogenous candidemia, both throughout
patients. The presence of a PICC was associated with the mucosal barrier damage and extensive exposure to broad-
greatest risk of candidemia at multivariate analysis: it inde- spectrum antibiotics or invasive medical devices, procedures
pendently increased the probability of candidemia more than and treatments, such as TPN [11, 34, 37, 38]. Of note, TPN
5 times. As previously shown by Tascini et al. [33], PICC is is a widely recognized risk factor for candidemia [39], and
a much stronger risk factor for candidemia than for bactere- it is a matter of fact that the presence of lipid emulsions and
mia, probably since it is a long standing device and mainly glucose in TPN formulations promotes the production of
predispose to the adherence of biofilm-producing yeasts Candida’s biofilm [40–42], making infusion lines and cath-
instead of bacteria. eters the ideal place for yeasts’ colonization and replication.
In our series, even other classical risk factors were related Moreover, prolonged fasting due to lack of enteral nutrition
to a significant increasing risk of candidemia. At multi- reduces physiological peristalsis and promotes alteration of
variate analysis CVC, TPN, previous or current antibiotic the intestinal barrier, thus facilitating the translocation into
therapy, and hospitalization within 3 months from the index bloodstream of intestinal microorganisms, including Can-
event, increased the risk of candidemia roughly two times dida spp [43].
with respect to bacteremia. All these are known risk factors Also Luzzati et al. [18] in a study on candidemia in IMWs
for candidemia and may reflect the frailty and complexity of found that risk factors significantly associated with candi-
the elderly people admitted to IMWs [31–33]. demia onset were the presence of a CVC and TPN with an
It is known that Candida species are an important cause increasing gradient sustained by the increasing duration of
of invasive infection at the extremes of age, and the elderly treatment.
population, which is increasing worldwide, is particularly More recently, Falcone et al. [44], studying risk factors
vulnerable due to high frequency of comorbidities, aging- for candidemia in a multicenter case–control study, identi-
related physiological changes, polypharmacy [34–36]. fied independent risk factors and derived a score that was
In the referral studies by Ostrosky-Zeichner and Leon on subsequently tested in a validation cohort. At multivariate
prediction rules for invasive candidiasis inside ICU and sur- analysis risk factors for candidemia resulted: Clostridium
gical settings, the mean age of enrolled patients was 59 and difficile infection, diabetes mellitus, TPN, chronic obstruc-
60 years, respectively [19, 20]. In our study, the mean age tive pulmonary disease (COPD), concomitant intravenous
of enrolled patients was 74 years with a median of 77 years, glycopeptide therapy, presence of PICC, previous antibiotic
closely reflecting current demographics of patients seen in therapy, and immunosuppressive therapy.
clinical practice. Our study, representing real-world general In our cohort, factors of increased risk of candidemia
medicine inpatients, can provide additional data on a very were quite similar, with the exception of C. difficile

13
 Author's personal copy
A prediction rule for early recognition of patients with candidemia in Internal Medicine: results…

infection (which was significant only at unadjusted analy- Although the limits of this work, we believe it could
sis), immunosuppressive therapy, glycopeptide therapy, help the internist physician in the challenge of early rec-
and comorbidities such as diabetes and COPD. ognition of a candidemia to start a timely empirical anti-
In our study, a factor significantly associated to the fungal therapy combined with empiric antibiotic therapy.
risk of candidemia was the presence of neurological dis- Infact, as shown in several studies, prompt diagnosis for a
ability, defined as stroke or dementia causing substantial timely treatment could be difficult in older patients (lack
functional impairment in daily activities. To the best of of fever or other clinical signs of systemic infection) [45].
our knowledge, this is the first time that neurological dis- But it is crucial in candidemic patients, being mortality
ability is shown as a risk factor for candidemia. Even if higher in case of a delay in antifungal treatment’s pre-
originally this observation is not surprising, since we can scription [3, 11]. This score could help to promptly start
argue that neurological disability caused by severe stroke empirical antifungal therapy in patients at risk for can-
or advanced dementia portrays a fragile subject with sub- didemia and could help physicians to select patients that
stantial loss of functional autonomy, often malnourished should undergo integrative non-cultural tests, such as
or at risk of malnutrition, needing supportive measures, mannan–antimannans and/or beta-d-glucan, in an appro-
and frequently administered through a PICC. Patients with priate manner for a better definition of the probability of
significant neurological disability often reside in long- candidemia. Also procalcitonin levels could be very use-
term care facilities, where the exposure to broad-spectrum ful to further discriminate between bacterial and fungal
antibiotics, a known risk factor for candidemia, is frequent bloodstream infections, but for the retrospective design
and may contribute to increasing the risk of candidemia. of our study, these data resulted lacking in many enrolled
The interpretation of the main differences among iden- patients, so we could not analyze it and incorporate into
tified risk factors in our study with respect to others may our score [46].
be various. We could hypothesize the different profile of This study has strengths and limitations. Among the
the hospitalized population of different Italian regions first ones, we underline the use of real-world data, the
and the different attitudes of different IMWs to a peculiar inclusion of patients with BSIs both in cases and con-
type of medical patients, that finally shows a different risk trols, the evaluation of a large number of predictors, the
toward candidemia (i.e., oncological, hematological, sub- availability of the information at admission time, and the
intensive, and post-surgical). conduction of sensitivity analyses. However, the study has
Importantly, the main difference among our series and some limitations as well. In clinical trials, randomization
the others is the selection of control cases. is intended to balance the distribution of both known and
Specifically, in our population both cases and controls unknown confounding factors in the compared groups.
are patients with BSIs, whereas sepsis was not a criterion This is rarely possible in observational studies and a
for selection of controls in the studies by Luzzati and Fal- systematic bias could be generated even when cases are
cone [18, 44]. In our opinion, this is a main and important matched with controls. Even if the 1:1 matching adopted
difference, since we deem fundamental trying to identify in this cohort could reflect the distribution of candidemia
risk factors for candidemia starting from the point of view and bacterial infections in the real practice, the scoring
of a potential septic syndrome, which constitute the starter system should be tested in a more balanced dataset. Selec-
for the initial clinical suspicion, as it was the cohort in tion bias might have occurred because patients with miss-
our study represented by patients with BSIs. This condi- ing data were excluded and the multivariate analysis was
tion could be a selection bias together with the fact that, restricted to patients who had complete data set. In addi-
primarily due the retrospective nature of the study, it is not tion, couples of variables were dichotomized. Although
possible to define exactly the grading of the septic state this strategy simplifies the creation of a risk score, the
of patients. This study identifies renal failure or cancer, use of continuous variables has the potential to provide
such as the presence of fever, treatments with steroids, more refined information. Despite a comparison with the
and Clostridium difficile infection, as risk factors for both existing test was performed, the smaller size of the cohort
candidemic and bacteriemic patients. We do not state that implies that these results should be reproduced in a larger
these risk factors do not play a major role as a risk factor one to achieve more robust results and compare to other
for infections due to Candida, but that does not allow us scoring systems. Finally, even if sensitivity analyses were
to identify the likelihood of a candidemia in a patient who conducted, the presence of a testing set should have made
has an infection. more rigorous the model validation process. For all these
We are aware that the prediction rule proposed requires reasons, further insights are needed and the proposed score
a prospective validation in a larger patient cohort, which should be used to drive perspective validations.
could possibly consider both bacterial and fungal sepsis and
negative controls.

13
 Author's personal copy
E. Sozio et al.

Conclusions Candidemia across five sites in Italy and Spain. J Clin Microbiol.
2013;51:4167–72.
11. Bassetti M, Taramasso L, Nicco E, Molinari MP, Mussap M, Vis-
The proposed scoring system obtained by the identification coli C. Epidemiology. Species distribution, antifungal suscepti-
of risk factors for candidemia in patients admitted to IMW bility and outcome of nosocomial Candidemia in a Tertiary Care
for bloodstream infections showed to be a simple and highly Hospital in Italy. PLoS One. 2011;6:e24198.
12. Luzzati R, Merelli M, Ansaldi F, et al. Nosocomial candidemia
performing tool in distinguishing candida and bacterial sep- in patients admitted to medicine wards compared to other wards:
sis through sensitivity analyses. The proposed rule might a multicentre study. Infection. 2016;44:747–55.
help to identify patients in IMWS at high risk of candidemia, 13. Ylipalosaari P, la-Kokko TI, Karhu J, et al. Comparison of the
guide the indication for non-cultural testing, reduce the delay epidemiology, risk factors, outcome and degree of organ failures
of patients with candidemia acquired before or during ICU treat-
in empirical treatment, and improve appropriateness in anti- ment. Crit Care. 2012;16(2):R62.
fungal prescription. Due to the retrospective nature of the 14. Milazzo L, Peri AM, Mazzali C, et al. Candidaemia observed
study, prospective validations in larger patient cohorts are at university hospital in Milan (northern Italy) and review
needed to confirm these preliminary findings. of published studies from 2010 to 2014. Mycopathologia.
2014;178:227–41.
15. De Rosa FG, Corcione S, Filippini C, et al. The effect on mortality
Funding  No financial support was received.
of fluconazole or echinocandin treatment in candidemia in internal
medicine wards. PLoS One. 2015;10:e0125149.
Compliance with ethical standards  16. Hii IM, Chang HL, Lin LC, et al. Changing epidemiology of
candidemia in a medical center in middle Taiwan. J Microbiol
Conflict of interest  CT has received funds for speaking at symposia Immunol Infect. 2015;48:306–15.
organized on behalf of Pfizer, Novartis, Merck, and Astellas. All other 17. Guimaraes T, Nucci M, Mendonca JS, et al. Epidemiology and
authors: none to declare. predictors of a poor outcome in elderly patients with candidemia.
Int J Infect Dis. 2012;16:e442–7.
18. Luzzati R, Cavinato S, Giangreco M, et al. Peripheral and total
parenteral nutrition as the strongest risk factors for nosocomial
candidemia in elderly patients: a matched case–control study.
References Mycoses. 2013;56:664–71.
19. León C, Ruiz-Santana S, Saavedra P, et al. A bedside scoring
system (“Candida score”) for early antifungal treatment in non-
1. Wisplinghoff H, Bischoff T, Tallent SM, Seifert H, Wenzel RP,
neutropenic critically ill patients with Candida colonization. Crit
Edmond MB. Nosocomial bloodstream infections in US hospitals:
Care Med. 2006;34:730–7.
analysis of 24,179 cases from a prospective nationwide surveil-
20. Ostrosky-Zeichner L, Sabel C, Sobel J, et al. Multicenter retro-
lance study. Clin Infect Dis. 2004;39:309–17.
spective development and validation of a clinical prediction rule
2. Marchetti O, Bille J, Fluckiger U, et al. Epidemiology of candidae-
for nosocomial invasive candidiasis in the intensive care setting.
mia in Swiss tertiary care hospitals: secular trends, 1991–2000.
Eur J Clin Microbiol Infect Dis. 2007;26:271–6.
Clin Infect Dis. 2004;38:311–20.
21. Paphitou NI, Ostrosky-Zeichner L, Rex JH. Rules for identify-
3. Morrell M, Fraser VJ, Kollef MH. Delaying the empiric treat-
ing patients at increased risk for candidal infections in the surgi-
ment of Candida bloodstream infection until positive blood culture
cal intensive care unit: approach to developing practical criteria
results are obtained: a potential risk factor for hospital mortality.
for systematic use in antifungal prophylaxis trials. Med Mycol.
Antimicrob Agents Chemother. 2005;49:3640–5.
2005;43:235–43.
4. Zimlichman E, Henderson D, Tamir O, et al. Health care-associ-
22. Pittet D, Monod M, Suter PM, Frenk E, Auckenthaler R. Candida
ated infections: a meta-analysis of costs and financial impact on
colonization and subsequent infections in critically ill surgical
the US health care system. JAMA Intern Med. 2013;173:2039–46.
patients. Ann Surg. 1994;220:751–8.
5. Cleveland AA, Farley MM, Harrison LH, et al. Changes in inci-
23. Dunning J, Au JK, Millner RW, Levine AJ. Derivation and valida-
dence and antifungal drug resistance in candidemia: results from
tion of a clinical scoring system to predict the need for an intra-
population-based laboratory surveillance in Atlanta and Balti-
aortic balloon pump in patients undergoing adult cardiac surgery.
more, 2008–2011. Clin Infect Dis. 2012;55:1352–61.
Interact Cardiovasc Thorac Surg. 2003;2:639–43.
6. Zaoutis TE, Argon J, Chu J, Berlin JA, Walsh TJ, Feudtner C.
24. Faiz S, Neale B, Rios E, et al. Risk-based fluconazole prophylaxis
The epidemiology and attributable outcomes of candidemia in
of Candida bloodstream infection in a medical intensive care unit.
adults and children hospitalized in the United States: a propensity
Eur J Clin Microbiol Infect Dis. 2009;28:689–92.
analysis. Clin Infect Dis. 2005;41:1232–9.
25. Dellinger RP, Levy MM, Rhodes A, et al. Surviving sepsis cam-
7. Wey SB, Mori M, Pfaller MA, Woolson RF, Wenzel RP. Hospital-
paign: international guidelines for management of severe sepsis
acquired candidemia. The attributable mortality and excess length
and septic shock. Intensive Care Med. 2013;39:165–228.
of stay. Arch Intern Med. 1988;148:2642–5.
26. Imtiaz R, Hawken S, McCormick BB, Leung S, Hiremath S, Zim-
8. Pfaller MA, Diekema DJ. Epidemiology of invasive candidi-
merman DL. Diabetes mellitus and younger age are risk factors
asis: a persistent public health problem. Clin Microbiol Rev.
for hyperphosphatemia in peritoneal dialysis patients. Nutrients.
2007;20:133–63.
2017;9:2.
9. Puig-Asensio M, Padilla B, Garnacho-Montero J, et al. Epidemiol-
27. Hon HH, Wojda TR, Barry N, et al. Injury and fatality risks
ogy and predictive factors for early and late mortality in Candida
in aeromedical transport: focus on prevention. J Surg Res.
bloodstream infections: a population-based surveillance in Spain.
2016;204:297–303.
Clin Microbiol Infect. 2014;20:O245–54.
28. Piombo AC, Gagliardi JA, Guetta J, et al. A new scoring sys-
10. Bassetti M, Merelli M, Righi E, et  al. Epidemiology, spe-
tem to stratify risk in unstable angina. BMC Cardiovasc Disord.
cies distribution, antifungal susceptibility, and outcome of
2003;3:8.

13
 Author's personal copy
A prediction rule for early recognition of patients with candidemia in Internal Medicine: results…

29. Soto A, Solari L, Agapito J, et al. Development of a clinical scor- 39. McClave SA, Martindale RG, Rice TW, Heyland DK. Feeding the
ing system for the diagnosis of smear-negative pulmonary tuber- critically ill patient. Crit Care Med. 2014;42:2600–10.
culosis. Braz J Infect Dis. 2008;12:128–32. 40. Shin JH, Kee SJ, Shin MG, et al. Biofilm production by isolates
30. Falagas ME, Aostolou KE, Pappas VD. Attributable mortality of of Candida species recovered from nonneutropenic patients: com-
candidemia. Eur J Clin Microbiol Infect Dis. 2006;25:419–25. parison of bloodstream isolates with isolates from other sources.
31. Peres-Bota D, Rodriguez-Villalobos H, Dimopoulos G, Melot C, J Clin Microbiol. 2002;40:1244–8.
Vincent JL. Potential risk factors for infection with Candida spp. 41. Branchini ML, Pfaller MA, Rhine-Chalberg J, Frempong T, Isen-
in critically ill patients. Clin Microbiol Infect. 2004;10:550–5. berg HD. Genotypic variation and slime production among blood
32. Wenzel RP. Nosocomial candidemia: risk factors and attributable and catheter isolates of Candida parapsilosis. J Clin Microbiol.
mortality. Clin Infect Dis. 1995;20:1531–4. 1994;32:452–6.
33. Tascini C, Sozio E, Tintori G, et al. Peripherally inserted central 42. Swindell K, Lattif AA, Chandra J, Mukherjee PK, Ghannoum
catheter as a predominant risk factor for candidemia in critically MA. Parenteral lipid emulsion induces germination of Candida
ill patients in internal medicine wards in Italy. Intensive Care Med. albicans and increases biofilm formation on medical catheter sur-
2015;41:1498–9. faces. J Infect Dis. 2009;200:473–80.
34. Luzzati R, Cavinato S, Deiana ML, Rosin C, Maurel C, Borelli M. 43. Pappo I, Polacheck I, Zmora O, Feigin E, Freund HR. Altered gut
Epidemiology and outcome of nosocomial candidemia in elderly barrier function to Candida during parenteral nutrition. Nutrition.
patients admitted prevalently in medical wards. Aging Clin Exp 1994;10:151–4.
Res. 2015;27:131–7. 44. Falcone M, Tiseo G, Tascini C, et al. Assessment of risk factors
35. Laupland KB, Gregson DB, Church DL, Ross T, Elsayed S. Inva- for candidemia in non-neutropenic patients hospitalized in internal
sive Candida species infections: a 5 year population-based assess- medical wards: a multicentre study. Eur J Int Med. 2017;41:33–8.
ment. J Antimicrob Chemother. 2005;56:532–7. 45. Tascini C, Falcone M, Bassetti M, et al. Candidemia in patient
36. Barchiesi F, Orsetti E, Mazzanti S, et  al. Candidemia in the with body temperature below 37 °C and admitted to internal medi-
elderly: what does it change? PLOS ONE. 2017;12:e0176576. cine wards: assessment of risk factors. Am J Med. 2016;129:1330.
37. Falcone M, Concia E, Iori I, et al. Identification and management e1–6.
of invasive mycoses in internal medicine: a road-map for physi- 46. Pieralli F, Corbo L, Torrigiani A, et al. Usefulness of procalcitonin
cians. Intern Emerg Med. 2014;9:501–11. in differentiating Candida and bacterial blood stream infections in
38. Tortorano AM, Prigitano A, Lazzarini C, et al. A 1-year prospec- critically ill septic patients outside the intensive care unit. Intern
tive survey of candidemia in Italy and changing epidemiology Emerg Med. 2017;12:629–35.
over one decade. Infection. 2013;41:655–62.

Affiliations

Emanuela Sozio1 · Filippo Pieralli2 · Anna Maria Azzini3 · Giancarlo Tintori4 · Federica Demma5 · Gianluca Furneri5 ·

Francesco Sbrana6 · Giacomo Bertolino7 · Simona Fortunato8 · Simone Meini9 · Damiano Bragantini3 ·
Alessandro Morettini10 · Carlo Nozzoli10 · Francesco Menichetti8 · Ercole Concia3 · Carlo Tascini11 · on behalf of GISA/
FADOI Candida Study Group

1 7
Emergency Department, North-West District, Tuscany Pharmaceutical Department, Ospedale di Sassuolo, Modena,
HealthCare, Spedali Riuniti Livorno, Viale Alfieri 36, Italy
57100 Leghorn, Italy 8
Infectious Diseases Clinic, Nuovo Santa Chiara University
2
Intermediate Care Unit, Azienda Ospedaliera Universitaria Hospital, Azienda Ospedaliera Universitaria Pisana, Pisa,
Careggi, Florence, Italy Italy
3 9
Infectious Disease Unit, Azienda Ospedaliera Universitaria Department of Internal Medicine, S.M. Annunziata Hospital,
Integrata di Verona, p.Le LA Scuro, Verona, Italy Florence, Italy
4 10
Emergency Medicine Unit, Nuovo Santa Chiara University Internal Medicine Unit, Azienda Ospedaliera Universitaria
Hospital, Azienda Ospedaliera Universitaria Pisana, Pisa, Careggi, Florence, Italy
Italy 11
First Division of Infectious Diseases, Cotugno Hospital,
5
Health Economics and Outcome Research Azienda Ospedaliera dei Colli, Naples, Italy
Department-EBMA Consulting, Milan, Italy
6
Fondazione Toscana Gabriele Monasterio, Pisa, Italy

13