Hirosawa et al.

BMC Infectious Diseases (2020) 20:85

https://doi.org/10.1186/s12879-020-4814-5

RESEARCH ARTICLE Open Access

Eosinopenia as a diagnostic marker of

bloodstream infection in a general internal
medicine setting: a cohort study
Takanobu Hirosawa, Yukinori Harada, Kohei Morinaga, Hiroshi Takase, Michihiro Nin and Taro Shimizu*

Abstract
Background: Little is known about the potential use of the eosinophil count as a predictive marker of bloodstream
infection. In this study, we aimed to assess the reliability of eosinopenia as a predictive marker of bloodstream
infection.
Methods: This retrospective cohort study was performed in the outpatient department and general internal
medicine department of a tertiary university hospital in Japan. A total of 189 adult patients with at least 2 sets of
blood cultures obtained during the period January 1–December 31, 2018, were included; those with the use of
antibiotic therapy within 2 weeks prior to blood culture, steroid therapy, or a history of haematological cancer were
excluded. The diagnostic accuracies of each univariate variable and the multivariable logistic regression models
were assessed by calculating the areas under the receiver operating characteristic curves (AUROCs). The primary
outcome was a positive blood culture indicating bloodstream infection.
Results: Severe eosinopenia (< 24.4 cells/mm3) alone yielded small but statistically significant overall predictive
ability (AUROC: 0.648, 95% confidence interval (CI): 0.547–0.748, P < 0.05), and only moderate sensitivity (68, 95% CI:
46–85%) and specificity (62, 95% CI: 54–69%). The model comprising baseline variables (age, sex), the C-reactive
protein level, and neutrophil count yielded an AUROC of 0.729, and further addition of eosinopenia yielded a slight
improvement, with an AUROC of 0.758 (P < 0.05) and a statistically significant net reclassification improvement (NRI)
(P = 0.003). However, the integrated discrimination index (IDI) (P = 0.284) remained non-significant.
Conclusions: Severe eosinopenia can be considered an inexpensive marker of bloodstream infection, although of
limited diagnostic accuracy, in a general internal medicine setting.
Keywords: Eosinopenia, Bloodstream infection, Blood culture

Background To date, no study has identified a highly sensitive and

Blood cultures are necessary for the diagnosis and man- specific, easily measured, rapid, and inexpensive marker
agement of patients with bloodstream infection [1]. How- of bloodstream infection that correlates with infection
ever, the usefulness of this test is limited except in special severity and prognosis. Although the presence of chills
situations [e.g. inpatients with suspected infectious endo- [7], the C-reactive protein (CRP) level [8, 9], and the
carditis [2] or meningitis [3]] because of its poor sensitivity quick Sequential (Sepsis-Related) Organ Failure Assess-
in ambulatory outpatient [4], primary care, and hospital ment (qSOFA) score [10] have been identified as poten-
internal medicine department settings [5]. Additionally, tial predictors of bloodstream infection, none has been
many contaminants may lead to a false positive culture determined to have adequate specificity and sensitivity.
and, consequently, unnecessary therapy [6]. Eosinopenia, defined as a reduced eosinophil count in
peripheral blood, was previously identified as a good
diagnostic marker of infection [11]. Although some stud-
* Correspondence: [email protected]
Department of Diagnostic and Generalist Medicine, Dokkyo Medical
ies reported that the absence of peripheral blood eosino-
University Hospital, Clinical Education Building, Kitakobayashi 880, Mibu, phils could not be used as a clinically reliable marker of
Shimotsuga, Mibu 321-0293, Japan

© The Author(s). 2020 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Hirosawa et al. BMC Infectious Diseases (2020) 20:85 Page 2 of 7

bacteraemia in a hospital inpatient setting [12, 13], those Patient population

studies included limited numbers of patients and were From a total of 399 adult patients (age > 15 years) who
not restricted to general internal medicine departments. underwent blood culture testing in the general internal
Therefore, the potential usefulness of eosinopenia as a medicine department during the study period, 205 were
predictor of bloodstream infection in patients presenting excluded because of antibiotic use within 2 weeks prior
or admitted to a general internal medicine department to the blood culture sampling (n = 178), steroid use (n =
remains unclear. In this study we hypothesised that eosi- 25), or haematological cancer (n = 2). Five other patients
nopenia would be a reliable marker of bloodstream in- were excluded because of a lack of data. The remaining
fection in adult patients treated in the general internal 189 patients were enrolled in the study. A flow diagram
medicine department of a tertiary university hospital. of patient selection is shown in Fig. 1. All blood cultures
were drawn at the discretion of the treating physician.
Methods
Study design and patient selection Patient and public involvement
This retrospective, single-centre cohort study included No patient involved.
all consecutive in- and outpatients in the general in-
ternal medicine department, excluding intensive care Outcome and definition
unit and emergency department, of Dokkyo Medical The primary study outcome was a positive blood culture
University Hospital, Mibu, Tochigi, Japan, who under- indicative of bloodstream infection. We defined a blood-
went blood culture testing from 1 January to 31 Decem- stream infection as the presence of a pathogenic micro-
ber, 2018. Dokkyo Medical University Hospital is a organism in at least one blood culture bottle. Samples
tertiary teaching hospital. This study was conducted in with bacterial contaminants were counted as negative
accordance with the current version of the Declaration cultures. The contamination criterion was the presence
of Helsinki. The study protocol was approved by the in- of multiplying coagulase-negative Staphylococcus species,
stitutional ethics committee of Dokkyo Medical Univer- Bacillus species, Propionibacterium acnes or Corynebac-
sity (No. R-20-18 J). terium species in a single set of blood cultures. These

Fig. 1 Flowchart of patient inclusion and exclusion in the study

Hirosawa et al. BMC Infectious Diseases (2020) 20:85 Page 3 of 7

bacteria were previously identified as frequent contami- Analysis

nants [14]. All such samples were excluded prior to the Continuous variables are presented as medians and
review of medical notes. interquartile ranges [25th–75th percentiles] and were
Absolute eosinopenia was differently defined in each re- compared using the Mann–Whitney U test. Categorical
search and does not have a universal definition [13, 15]. In or binary variables are presented as numbers (percent-
this study, the optimal cut-off was defined as an eosinophil ages) and were compared using the chi-squared test or
count of < 24.3 cells/mm3 from univariate analysis. The Fisher’s exact test. The diagnostic accuracies of each uni-
qSOFA, a recently developed measure for the rapid identi- variate variable and the multivariable logistic regression
fication of infected patients at risk of mortality, was also models were assessed by calculating the corresponding
applied [10, 16, 17]. This bedside clinical score identifies area under each receiver operating characteristic curve
adult patients with suspected infection and a higher risk of (AUROC). A P value of < 0.05 was considered statisti-
poor outcomes typical of sepsis as those who meet with at cally significant. The 95% confidence intervals (CIs) were
least 2 of the following clinical criteria: respiratory rate of used to quantify uncertainty.
≥22/min, altered mentation, or systolic blood pressure of Previous studies identified the CRP level as a powerful
≤100 mmHg [10]. predictive marker of bloodstream infection [11, 20]. In
this study, we calculated the integrated discrimination
index (IDI) and net reclassification improvement (NRI)
Procedure
[21] to assess whether the inclusion of eosinopenia into
From each patient, the clinicians drew 10 mL of blood
the model involving the baseline variables (age + sex)
aseptically from a superficial vein and inoculated the
and CRP level would improve the predictive value. All
sample into both aerobic and anaerobic cultures. They
statistical tests were performed using the R 3.6.0 and
repeated the procedure using a different superficial vein
pROC package [22] for MacOS X (The R foundation for
to yield 2 sets of blood cultures for each patient [18].
Statistical Computing, Vienna, Austria). Internal valid-
The cultures were incubated in blood culture bottles
ation of the prediction models was conducted using or-
containing BACTEC resin-beads (Bactec Plus Aerobic/
dinary nonparametric bootstrapping with 1000 bootstrap
23F and Anaerobic/22F bottles; Becton Dickinson In-
samples and bias-corrected, accelerated 95% CIs [23].
strument Systems, Sparks, MD, USA). The bottles were
incubated at 35 °C, sub-cultured daily, and inspected for
Results
bacterial growth for 6 days.
Of the 189 patients enrolled in the final analysis, 25 and
A fully automated BACTEC-FX blood culture incuba-
80 patients with a positive blood culture or eosinopenia,
tion system (Becton Dickinson) was used to isolate bac-
respectively, were identified during the study period. In
teria from the blood cultures. Significant isolates were
4 patients, multiple organisms were detected in the same
identified and tested for antimicrobial susceptibility ac-
blood culture specimen at the time of bloodstream infec-
cording to the National Committee for Clinical Labora-
tion diagnosis; 12 of the 25 identified bloodstream infec-
tory Standards guidelines [19]. All bacterial species
tions (48%) were due to Gram-positive organisms, while
isolated from blood culture bottles were confirmed using
9 (36%) were due to Gram-negative organisms. The
matrix-assisted laser desorption ionisation–time of flight
baseline characteristics of infected and non-infected pa-
mass spectrometry.
tients are shown in Table 1. Patients with a bloodstream
infection had a significantly higher total white cell count
Data collection and CRP concentration than those without a blood-
Patients’ medical records were reviewed to ensure that 2 stream infection. All other comparisons yielded insignifi-
attending clinicians considered the detected micro- cant results. Other patient characteristics that might
organisms to be pathological, rather than contaminants. have affected the eosinophil count [15] are presented in
All data in this study were collected by the treating clini- the Additional file 1: Table.S1.
cians in the context of clinical management and in- Table 2 presents the results of univariate analyses.
cluded age, sex, presence of chills, and vital signs Eosinopenia (AUROC: 0.648, 95% CI: 0.547–0.748, cut-
(mental status, respiratory rate, and systolic blood pres- off = 24.4 cells/mm3), neutrophil count (AUROC: 0.638,
sure) at the time of blood culture sampling. The poten- 95% CI: 0.519–0.758, cut-off = 9033 cells/mm3), and
tial markers of bloodstream infection assessed in this CRP concentration (AUROC: 0.699, 95% CI: 0.597–
study included the serum CRP concentration and total 0.802, cut-off = 4.89 mg/dL) were all identified as signifi-
white blood cell, neutrophil, and eosinophil count. All cant predictive markers of bloodstream infection. Fol-
markers were measured within 1 day of blood culture lowing bootstrapped multiple regression analysis (1000
sample collection. Eosinophil count was determined bootstrap replicates), eosinopenia showed the same
using an automated method. AUROC and 95% CI (AUROC: 0.648, 95% CI: 0.557–
Hirosawa et al. BMC Infectious Diseases (2020) 20:85 Page 4 of 7

Table 1 Comparison of characteristics between patients with and without bloodstream infection
Variable Bloodstream infection No bloodstream infection(n = 164) P value*
(n = 25)
Age, y (SD) [median] 71.8 (15.8) [75.0] 62.8 (20.0) [68.0] 0.246
Male, n (%) 12 (46) 93 (57) 0.582
CRP, mg/l [median, IQR] 120.5 [93.8, 50.4–160.7] 63.5 [40.4, 8.7–88.7] 0.017
3
Total white cell count, cells/mm [median, IQR] 11,360 [11,100, 8400-12,700] 9901 [8900, 6700-12,000] 0.009
Eosinophil count, cells/mm3 [median, IQR] 32.0 [11.0, 0.00–38.4] 115.1 [37.6, 0.00–100.3] 0.741
Neutrophil count, cells/mm3 [median, IQR] 10,141 [9601, 6969-11,842] 7971 [7250, 4754-9964] 0.075
qSOFA score 0–1 22 145 0.208
qSOFA score 2–3 3 19
Chills, n (%) 8 (32) 34 (20) 0.891
SD standard deviation, IQR interquartile range, CRP C-reactive protein. QSOFA Quick Sequential (Sepsis-Related) Organ Failure Assessment
*P values by chi-squared, Mann-Whitney U test, or Fisher’s exact test

0.743). In contrast, white cell count (P = 0.185), qSOFA Discussion

(P = 0.502), and presence of chills (P = 0.211) were not According to our findings, eosinopenia alone yielded a
identified as statistically significant predictive markers. reasonable overall predictive ability, but only moderate
Further analysis revealed that eosinopenia could predict sensitivity and specificity for bloodstream infection in a
bloodstream infection with only moderate specificity (62, cohort of patients who presented or were admitted to
95% CI: 54–69%) and sensitivity (68, 95% CI: 46–85%). the department of general internal medicine at our uni-
The CRP concentration was more sensitive (80, 95% CI: versity hospital. However, we found that eosinopenia
61–93%) but less specific (56, 95% CI: 48–64%). The was a more useful predictor of bloodstream infection
neutrophil count was less sensitive (61, 95% CI: 41–80%) than the qSOFA score and presence of chills in general
but more specific (69, 95% CI: 62–77%). internal medicine setting, excluding intensive care unit
Table 3 presents the AUROCs of the predictive models and emergency department. Moreover, the inclusion of
for bloodstream infection. The addition of CRP and neu- eosinopenia in the prediction model comprising the
trophil count to the baseline variables (age, sex) im- baseline variables and CRP led to a slight improvement
proved the AUROC (from 0.650 to 0.729; P = 0.002) and in the AUROC. These results suggest that eosinopenia
yielded a statistically significant IDI (P = 0.023) and NRI may be useful as an inexpensive predictor of blood-
(P = 0.005). Further addition of eosinopenia to the model stream infections. However, further investigations would
including the baseline variables, CRP, and neutrophil be needed to exclude bloodstream infection.
count led to a slight improvement, with an AUROC of Our study can be distinguished from previous work by
0.758 (P = 0.048) and a statistically significant NRI (P = a notable strength, namely the collection of data from a
0.003). However, the IDI (P = 0.284) was not significant. general internal medicine department. Although chills or
Following bootstrapped multiple regression analysis, the qSOFA were previously identified as useful predictors of
model with eosinopenia showed the same AUROC. The bloodstream infection in an intensive care unit or emer-
corresponding ROC curves are shown in Fig. 2. gency department setting [11, 20, 24], our study showed

Table 2 Areas under the receiver operating characteristic curves of eosinophil, total white cell, neutrophil count, CRP, and qSOFA as
potential markers of bloodstream infection identified through univariate analysis
Variable Cut-off value AUROC (95% CI) P value*
Eosinophil count < 24.4 cells/mm3 0.648 (0.547–0.748)0.648 (0.557–0.743)** 0.007
3
White cell count > 10,950 cells/mm 0.597 (0.472–0.723) 0.185
Neutrophil count > 9033 cells/mm3 0.638 (0.519–0.758) 0.040
CRP, mg/l > 4.89 mg/dl 0.699 (0.597–0.802) 0.001
qSOFA 0.502 (0.433–0.572) 0.952
Chills 0.556 (0.458–0.655) 0.211
AUROC Area under the receiver operating characteristic curve
CI confidence interval, CRP C-reactive protein
QSOFA Quick Sequential (Sepsis-Related) Organ Failure Assessment
*P values by chi-squared, Mann-Whitney U test, or Fisher’s exact test
**Bootstrapping method (1000 bootstrap replicates)
Hirosawa et al. BMC Infectious Diseases (2020) 20:85 Page 5 of 7

Table 3 Areas under the receiver operating characteristic curves of the predictive models for bloodstream infection
Model AUROC (95% CI) P value IDI P value NRI P value
Baseline variables* 0.650 (0.551–0.749)
Baseline variables 0.729 (0.622–0.835) 0.002** 0.069** 0.023** 0.583** 0.005**
+ CRP
+ neutrophil count
Baseline variables 0.758 0.048*** 0.016*** 0.284*** 0.592*** 0.003***
+ CRP (0.664–0.853)
+ neutrophil count 0.758
+ eosinopenia (0.667–0.845)****
AUROC Area under the receiver operating characteristic curves
CI confidence interval, CRP C-reactive protein
IDI integrated discrimination index
NRI net reclassification improvement
* Including age, sex
** Compared with the model with baseline variables
***Compared with the model with baseline variables + CRP + neutrophil count
****Bootstrapping method (1000 bootstrap replicates)

that neither factor was a significant predictor of blood- an elevated CRP level and elevated neutrophil count to
stream infection in our general internal medicine depart- the baseline variables yielded a stronger predictive meas-
ment. This inconsistency suggests that chills and qSOFA ure. Further addition of severe eosinopenia to this model
may only be useful predictors in patients with a severe also led to a slight improvement in the predictive ability,
acute condition, who often present to an emergency de- even though the eosinophil count itself was not a suffi-
partment or are admitted to intensive care unit, but not ciently predictive marker.
in patients with milder conditions who would present to This study had several limitations. First, it was con-
a general internal medicine department. Additionally, ducted in a single department at a single centre, and
our finding that eosinopenia is a predictor of blood- therefore, our results cannot be easily generalised to the
stream infection suggests that this marker may be a use- intensive care unit, surgery, and emergency department.
ful tool for predicting such infections in patients with a Second, we excluded 178 patients (44.6%) who used an-
mild general condition. We found that the addition of tibiotics within 2 weeks prior to blood culture sampling,

Fig. 2 Areas under the receiver operating characteristic curves associated with bloodstream infection for the baseline variables alone (black line),
baseline variables + C-reactive protein + neutrophil count (CRP, blue line), and baseline variables + CRP + neutrophil count + eosinopenia (red line)
Hirosawa et al. BMC Infectious Diseases (2020) 20:85 Page 6 of 7

which may pose a risk of inducing a selection bias. This Availability of data and materials
may be due to the tertiary nature of our institution, as The datasets generated and/or analysed in this study are not publicly
available because it is possible that individual anonymity could be
patients with bloodstream infection may have initially compromised.
visited a primary clinic and began to receive antibiotic
treatment prior to referral to our department. Third, no Ethics approval and consent to participate
clear criteria have been set to determine which patients Ethics committee of Dokkyo Medical University (No. R-20-18 J). The commit-
tee approved participation of patients was obtained through an opt-out
would be subjected to blood culture. Rather, this deci- methodology. The patients consented to analysis of their medical records
sion was made by the treating physician on a case-by- was not required. Any further permission from the hospital was not required.
case basis. Fourth, we excluded patients with haemato- This was because of all data was obtained from usual clinical process.
logical diseases, eosinophilia, and steroid users, as such
Consent for publication
cases are rarely seen in our department. Accordingly, Not applicable.
our findings are not generalisable to these patient groups
or areas with a high prevalence of these diseases. Fifth, Competing interests
not only severe and shaking but also mild to moderate None declared.
chills were included in this study. According to the ori-
Received: 29 July 2019 Accepted: 21 January 2020
ginal article, the more severe degree of chills, especially
shaking chills, suggests the high risk of bacteremia [7].
That could explain our insignificant results. Finally, pro- References
calcitonin was recently identified as a novel predictive 1. Kirn TJ, Weinstein MP. Update on blood cultures: how to obtain, process,
report, and interpret. Clin Microbiol Infect. 2013;19(6):513–20.
marker of bloodstream infection, although one study 2. Baddour LM, Wilson WR, Bayer AS, Fowler VG Jr, Tleyjeh IM, Rybak MJ, et al.
identified it as a poor predictor of culture positivity [25]. Infective endocarditis in adults: diagnosis, antimicrobial therapy, and
In our study, procalcitonin was not evaluated in most Management of Complications: a scientific statement for healthcare
professionals from the American Heart Association. Circulation. 2015;132(15):
cases. The results of our AUROC analysis suggest that 1435–86.
when eosinopenia was modelled as a continuous vari- 3. Tunkel AR, Hartman BJ, Kaplan SL, Kaufman BA, Roos KL, Scheld WM, et al.
able, its diagnostic utility as a marker of bloodstream in- Practice guidelines for the management of bacterial meningitis. Clin Infect
Dis. 2004;39(9):1267–84.
fection was limited. 4. Laupland KB, Church DL, Gregson DB. Blood cultures in ambulatory
outpatients. BMC Infect Dis. 2005;5:35.
5. Friedman ND, Braun T, Fallach N, Carmeli Y. Blood Culture Sampling
Conclusion Practices among Internal Medicine Inpatients. Clin Microbiol Infect Dis. 2017;
In summary, severe eosinopenia can be considered an 2(1):1-6.
inexpensive marker of bloodstream infection, although 6. Bates DW CE, Goldman L, Lee TH. Predicting bacteremia in hospitalized
patients. A prospectively validated model. Annals of internal medicine. 1990;
of limited diagnostic accuracy, in a general internal 113(7):6.
medicine setting. 7. Tokuda Y, Miyasato H, Stein GH, Kishaba T. The degree of chills for risk of
bacteremia in acute febrile illness. Am J Med. 2005;118(12):1417.
8. Black S, Kushner I, Samols D. C-reactive protein. J Biol Chem. 2004;279(47):
Supplementary information 48487–90.
Supplementary information accompanies this paper at https://doi.org/10. 9. Mccabe RE, Remington JS. C-reactive protein in patients with bacteremia. J
1186/s12879-020-4814-5. Clin Microbiol. 1984;20(3):317–9.
10. Singer M, Deutschman CS, Seymour CW, Shankar-Hari M, Annane D, Bauer
Additional file 1. Additional characteristics of patients who underwent M, et al. The third international consensus definitions for Sepsis and septic
blood culture. shock (Sepsis-3). Jama. 2016;315(8):801.
11. Abidi K, Khoudri I, Belayachi J, Madani N, Zekraoui A, Zeggwagh AA, et al.
Eosinopenia is a reliable marker of sepsis on admission to medical intensive
Abbreviations care units. Crit Care. 2008;12(2):R59.
AUROCs: Areas under the receiver operating characteristic curves; CRP: C- 12. Setterberg MJ, Newman W, Potti A, Smego RA Jr. Utility of eosinophil count
reactive protein; IDI: Integrated discrimination index; NRI: Net reclassification as predictor of bacteremia. Clin Infect Dis. 2004;38(3):460–1.
improvement; QSOFA: quick Sequential (Sepsis-Related) Organ Failure 13. Wibrow BA, Ho KM, Flexman JP, Keil AD, Kohrs DL. Eosinopenia as a
Assessment diagnostic marker of bloodstream infection in hospitalised paediatric and
adult patients: a case-control study. Anaesth Intensive Care. 2011;39(2):
224–30.
Acknowledgements
14. Terradas R, Grau S, Blanch J, Riu M, Saballs P, Castells X, et al. Eosinophil
Not applicable.
count and neutrophil-lymphocyte count ratio as prognostic markers in
patients with bacteremia: a retrospective cohort study. Plos One. 2012;7(8):
Authors’ contributions e42860.
TH, YH, KM, HT, MN and TS contributed to the study concept and design. TH 15. Karakonstantis S, Kalemaki D, Tzagkarakis E, Lydakis C. Pitfalls in studies of
and YH performed the statistical analyses. TH contributed to the drafting of eosinopenia and neutrophil-to-lymphocyte count ratio. Infectious diseases.
the manuscript. YH, KM, HT, MN and TS contributed to the critical revision of 2018;50(3):163–74.
the manuscript for important intellectual content. All authors read and 16. Seymour CW, Gesten F, Prescott HC, Friedrich ME, Iwashyna TJ, Phillips GS,
approved the final manuscript. et al. Time to treatment and mortality during mandated emergency Care
for Sepsis. N Engl J Med. 2017;376(23):2235–44.
Funding 17. Ferrer R, Martin-Loeches I, Phillips G, Osborn TM, Townsend S, Dellinger RP,
No funding was obtained for this study. et al. Empiric antibiotic treatment reduces mortality in severe sepsis and
Hirosawa et al. BMC Infectious Diseases (2020) 20:85 Page 7 of 7

septic shock from the first hour: results from a guideline-based performance
improvement program. Crit Care Med. 2014;42(8):1749–55.
18. Weinstein MP. Current blood culture methods and systems: clinical
concepts, technology, and interpretation of results. Clin Infect Dis. 1996;
23(1):40–6.
19. Wayne P. CLSI. Performance Standards for Antimicrobial Susceptibility
Testing CLSI Supplement M100S. 26th ed. Clinical and Laboratory Standards
Institute; 2016.
20. Shaaban H, Daniel S, Sison R, Slim J, Perez G. Eosinopenia: is it a good
marker of sepsis in comparison to procalcitonin and C-reactive protein
levels for patients admitted to a critical care unit in an urban hospital? J Crit
Care. 2010;25(4):570–5.
21. Pencina MJ, D'Agostino RB Sr, D'Agostino RB Jr, Vasan RS. Evaluating the
added predictive ability of a new marker: from area under the ROC curve to
reclassification and beyond. Stat Med. 2008;27(2):157–72 discussion 207-12.
22. Robin X, Turck N, Hainard A, Tiberti N, Lisacek F, Sanchez JC, et al. pROC: an
open-source package for R and S+ to analyze and compare ROC curves.
BMC Bioinformatics. 2011;12:77.
23. Efron B TR. An introduction to the bootstrap: CRC press; 1994.
24. Lavoignet CE, Le Borgne P, Slimani H, Forato M, Kam C, Kauffmann P, et al.
Relevance of eosinopenia as marker of sepsis in the emergency
department. Rev Med Interne. 2016;37(11):730–4.
25. Caffarini EM, DeMott J, Patel G, Lat I. Determining the Clinical Utility of an
Absolute Procalcitonin Value for Predicting a Positive Culture Result.
Antimicrob Agents Chemother. 2017;61, 5.

Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.