Use of a Combination Biomarker Algorithm To Identify Medical

Intensive Care Unit Patients with Suspected Sepsis at Very Low

Likelihood of Bacterial Infection
Jennifer H. Han,a,b,c Irving Nachamkin,d Susan E. Coffin,f,g Jeffrey S. Gerber,b,c,f,g Barry Fuchs,e Charles Garrigan,d Xiaoyan Han,a,b
Warren B. Bilker,b,c Jacqueleen Wise,b Pam Tolomeo,b Ebbing Lautenbach,a,b,c for the Prevention Epicenters Program of the Centers
for Disease Control and Prevention
Division of Infectious Diseases, Department of Medicine,a Center for Clinical Epidemiology and Biostatistics,b Department of Biostatistics and Epidemiology,c Department
of Pathology and Laboratory Medicine,d and Division of Pulmonary and Critical Care Medicine, Department of Medicine,e Perelman School of Medicine, University of

Downloaded from http://aac.asm.org/ on December 11, 2019 by guest

Pennsylvania, Philadelphia, Pennsylvania, USA; Division of Infectious Diseasesf and Center for Pediatric Clinical Effectiveness,g Children’s Hospital of Philadelphia,
Philadelphia, Pennsylvania, USA

Sepsis remains a diagnostic challenge in the intensive care unit (ICU), and the use of biomarkers may help in differentiating bac-
terial sepsis from other causes of systemic inflammatory syndrome (SIRS). The goal of this study was to assess test characteristics
of a number of biomarkers for identifying ICU patients with a very low likelihood of bacterial sepsis. A prospective cohort study
was conducted in a medical ICU of a university hospital. Immunocompetent patients with presumed bacterial sepsis were con-
secutively enrolled from January 2012 to May 2013. Concentrations of nine biomarkers (␣-2 macroglobulin, C-reactive protein
[CRP], ferritin, fibrinogen, haptoglobin, procalcitonin [PCT], serum amyloid A, serum amyloid P, and tissue plasminogen acti-
vator) were determined at baseline and at 24 h, 48 h, and 72 h after enrollment. Performance characteristics were calculated for
various combinations of biomarkers for discrimination of bacterial sepsis from other causes of SIRS. Seventy patients were in-
cluded during the study period; 31 (44%) had bacterial sepsis, and 39 (56%) had other causes of SIRS. PCT and CRP values were
significantly higher at all measured time points in patients with bacterial sepsis. A number of combinations of PCT and CRP,
using various cutoff values and measurement time points, demonstrated high negative predictive values (81.1% to 85.7%) and
specificities (63.2% to 79.5%) for diagnosing bacterial sepsis. Combinations of PCT and CRP demonstrated a high ability to dis-
criminate bacterial sepsis from other causes of SIRS in medical ICU patients. Future studies should focus on the use of these al-
gorithms to improve antibiotic use in the ICU setting.

A ntibiotic resistance represents a major public health chal-

lenge, and strategies to promote the judicious use of antibiot-
ics are urgently needed. The intensive care unit (ICU) is a health
care setting in which rates of antibiotic use are particularly high (1,
2). Given the diagnostic challenges in this critically ill patient pop-
ulation, the use of empirical broad-spectrum antibiotics is fre-
quently initiated (or the range of antibiotics used is broadened) in
the presence of systemic inflammatory syndrome (SIRS) and of-
ten continued for prolonged courses despite the absence of clinical
and microbiological data supporting a diagnosis of bacterial infec-
tion. Inappropriate or excessive antibiotic exposure leads to the
emergence of bacterial resistance, increased health care costs, and
antibiotic-associated adverse events, including adverse drug
events and Clostridium difficile infection (3, 4).
Several studies have evaluated the diagnostic utility of various
biomarkers, including ferritin, haptoglobin, interleukin 6, C-re-
active protein (CRP), and procalcitonin (PCT), for suspected sep-
sis in the ICU patient population (5–15). PCT has been the
most-well-studied biomarker and has demonstrated the most
promising results to date in the diagnosis of bacterial infection in
a number of settings (16–21). However, recent meta-analyses have
suggested that PCT alone may be insufficient to accurately differ-
entiate bacterial from nonbacterial sepsis or other causes of SIRS
(22–24). Similarly, previous studies assessing the use of PCT alone
or in comparison to other biomarkers for the diagnosis of sepsis
have had variable performance characteristics (5–15), likely due to
differences in patient populations, the type of assay performed,
the timing of sample testing, the selection of cutoff values, and the
selection of the gold standard for determining sepsis.
While severe bacterial infections typically trigger the produc-
tion of multiple cytokines and other inflammatory proteins (25),
very few studies have assessed the utility of a combination of bio-
markers in the diagnosis of bacterial sepsis. The measurement of
multiple biomarkers may provide a more accurate strategy to di-
agnose bacterial sepsis in the ICU setting. In addition, for the
critically ill patient population in whom timely administration of
empirical broad-spectrum antibiotics significantly reduces mor-
tality, the greatest clinical utility of biomarkers may be to support
Received 22 April 2015 Returned for modification 2 June 2015
Accepted 30 July 2015
Accepted manuscript posted online 3 August 2015
Citation Han JH, Nachamkin I, Coffin SE, Gerber JS, Fuchs B, Garrigan C, Han X,
Bilker WB, Wise J, Tolomeo P, Lautenbach E, for the Prevention Epicenters
Program of the Centers for Disease Control and Prevention. 2015. Use of a
combination biomarker algorithm to identify medical intensive care unit patients
with suspected sepsis at very low likelihood of bacterial infection. Antimicrob
Agents Chemother 59:6494 –6500. doi:10.1128/AAC.00958-15.
Address correspondence to Jennifer H. Han, jennifer.han@uphs.upenn.edu.
Supplemental material for this article may be found at http://dx.doi.org/10.1128
/AAC.00958-15.
Copyright © 2015, American Society for Microbiology. All Rights Reserved.
doi:10.1128/AAC.00958-15

6494 aac.asm.org Antimicrobial Agents and Chemotherapy October 2015 Volume 59 Number 10

clinical decisions to discontinue antibiotics when no bacterial in-
fection is identified. Therefore, we conducted this study to assess
test characteristics of a number of candidate biomarkers for iden-
tifying critically ill patients in the ICU setting with a very low
likelihood of bacterial sepsis.
MATERIALS AND METHODS
Study design and setting. A prospective cohort study was conducted in
the 24-bed adult medical ICU (MICU) of the hospital of the University of
Pennsylvania, a quaternary care academic medical center. The initial
source population included all patients located in the MICU from January
2012 to May 2013.
Study population. A patient was considered potentially eligible for the
study if, on ICU admission or at any time during the ICU stay, the patient
was identified as having presumed bacterial sepsis. Specifically, patients
were required to meet established criteria for SIRS as well as have new
empirical antibiotic therapy initiated and blood cultures ordered, thereby
indicating the suspicion of bacterial infection (26, 27). Daily screening of
all MICU patient orders was performed using the hospital computerized
order entry system (Sunrise Clinical Manager; Allscripts, Chicago, IL).
SIRS was considered present when patients met at least two of the four
following criteria within 4 h of the enrollment blood culture collection
time: body temperature of ⬎38°C or ⬍36°C, heart rate of ⬎90/min, re-
spiratory rate of ⬎20/min, or a white blood cell count (WBC) of ⬎12,000
cells/␮l or ⬍4,000/␮l. New empirical antibiotic therapy was defined as
either the initiation of new antibiotic therapy in a patient previously not
on any antibiotics or the broadening of antibiotic therapy (determined by
an infectious diseases-trained physician, E.L.) in a patient who was already
receiving at least one antibiotic.
Subsequently, patients were considered eligible for enrollment unless
one of the following exclusion criteria was present: (i) a code status of do
not resuscitate; (ii) cardiopulmonary arrest, with the requirement for re-
suscitation; (iii) initiation of broad-spectrum antibiotic therapy for a doc-
umented bacterial infection in the 5 days prior to enrollment, as the algo-
rithm would be of low utility in these patients (e.g., antibiotic therapy
would not be discontinued, regardless of biomarker results); and (iv) the
presence of an immunocompromising condition, including human im-
munodeficiency virus (HIV) infection with a CD4 count ⬍200 cell/mm3,
immunosuppressive therapy after organ transplantation, neutropenia
(⬍500 neutrophils/mm3), chemotherapy, receipt of ⱖ20 mg/day of pred-
nisone for ⱖ2 weeks in the preceding 3 months, or cystic fibrosis.
Last, patients were excluded if they were receiving new or broadened
empirical antibiotic therapy for ⬎4 h past the time point when baseline
biomarkers were measured, given the potential effect of antibiotic therapy
on decreasing PCT values (28).
Biomarker measurements. Blood samples for biomarker measure-
ments were obtained from residual blood samples of tests performed dur-
ing routine clinical care. Candidate biomarkers were measured at the time
when a patient met all eligibility criteria (baseline) and then daily for 3
days (at 24 h, 48 h, and 72 h).
The timeline range for selecting a residual blood sample for biomarker
measurements was as follows: baseline, i.e., 12 h before empirical antibi-
otics were started or broadened to 4 h after; 24 h, i.e., 4 h after empirical
antibiotics were started or broadened to 36 h after the baseline sample; 48
h, i.e., 36 h after the baseline sample to 60 h after the baseline sample; and
72 h, i.e., 60 h after the baseline sample to 90 h after the baseline sample. If
there were two or more residual blood samples in the timeline range, the
one closest to the exact time point of interest was selected.
Nine candidate biomarkers were measured at each time point (Table
1). PCT levels were measured using the Vidas Brahms PCT assay
(bioMérieux, Durham, NC), a one-step immunoassay sandwich method
with fluorescent detection. The remaining eight biomarkers were mea-
sured using a Bio-Plex Pro human acute phase 5-plus-4-plex panel com-
plete kit (Bio-Rad Laboratories, Hercules, CA). This assay is a bead-based
(xMAP technology) multiplex assay that allows for the simultaneous mea-
October 2015 Volume 59 Number 10 Antimicrobial Agents and Chemotherapy
Biomarkers and Sepsis Diagnosis
TABLE 1 Candidate biomarkers
Biomarker Assay range, ng/ml
␣-2 Macroglobulin 0.5–1,875
C-reactive protein 0.01–50
Ferritin 3.06–50,000a
Fibrinogen 5–813
Haptoglobin 0.1–500
Procalcitonin 0.05–200
Serum amyloid A 1–700
Serum amyloid P 0.1–250
Tissue plasminogen activator 28–5,000a
a
Values for ferritin and tissue plasminogen activator are in pg/ml.
Downloaded from http://aac.asm.org/ on December 11, 2019 by guest
surement of nine positive acute-phase biomarkers in serum. The assay was
performed as per the manufacturer’s instruction and using a Luminex 200
reader (Luminex Corporation, Austin, TX). At the time of measurement,
serum samples from all four time points for each patient were included in
the same measurement test run. Each analyte was measured in duplicate,
as suggested by the manufacturer. A single lot was used for all measure-
ments with the Bio-Rad assay, and results were recorded as the mean for
each measurement.
The study was approved by the institutional review board of the Uni-
versity of Pennsylvania. Because routine blood tests are performed at least
twice a day in the MICU for patient care, and because residual blood
volumes from these samples are held in the central receiving laboratory,
the requirement for informed consent was waived.
Data collection. Variables recorded at baseline included demographic
information (e.g., age, race), comorbidities, and length of hospital and
MICU stay prior to enrollment. Comorbidities included hepatic dysfunc-
tion, malignancy (hematologic and solid tumor), diabetes mellitus,
chronic kidney disease (including the requirement for hemodialysis), and
pulmonary disease (i.e., chronic obstructive pulmonary disease or chronic
bronchitis). Hepatic dysfunction was defined as ⱖ2 of the following: total
bilirubin concentration of ⬎2.5 mg/dl; aspartate aminotransferase or al-
anine aminotransferase level more than twice the upper normal range; or
documented diagnosis of cirrhosis. Finally, the acute physiology and
chronic health evaluation (APACHE) II score was calculated for all pa-
tients upon enrollment (29).
Definition of infection. At 72 h, the definitive diagnosis of bacterial
infection was determined by the investigators using established Centers
for Disease Control and Prevention criteria (30). Medical record review,
including vital signs, provider notes, and laboratory and radiographic
results, was performed by two infectious diseases-trained physicians (E.L.
and J.H.H.) who were blinded to the results of biomarkers testing. Dis-
cordant results were adjudicated by further discussion between the two
investigators. This assessment of bacterial infection at 72 h served as the
gold standard for determining the test characteristics of the biomarkers.
Statistical analysis. Descriptive statistics, including comparisons be-
tween patients with bacterial sepsis versus other causes of SIRS (i.e., non-
bacterial sepsis or noninfectious SIRS), were calculated to characterize the
enrolled MICU patient population. Continuous variables were compared
using the Wilcoxon rank sum test, and categorical variables were com-
pared using Fisher’s exact test. Subsequently, the values of each biomarker
were plotted for patients with bacterial sepsis versus those with other
causes of SIRS and assessed visually to explore various cutoff values. Cut-
off values were then selected for biomarkers that appeared to most opti-
mally discriminate between patients with bacterial sepsis and those with
other causes of SIRS. Performance characteristics (e.g., negative predictive
value [NPV], specificity) were determined for various combinations of
time points (24 h, 48 h, 72 h) and cutoff values for these biomarkers. The
discriminatory ability of the biomarker algorithm was quantified using
the C statistic or the area under the receiver-operating characteristic
(ROC) curve (AUC). ROC curves were generated for individual biomark-
ers as well as for combinations of biomarkers and their interaction.
aac.asm.org 6495
Han et al.
TABLE 2 Characteristics among medical intensive care unit patients with bacterial sepsis and other causes of SIRS
Variable Bacterial sepsis (n ⫽ 31)
Mean (SD) age (yr) 60.8 (18.4)
No. (%) non-white race 17 (55)
No. (%) female sex 8 (26)
Mean (SD) ICU length of stay prior to enrollment (days) 1.7 (2.9)
Mean (SD) hospital length of stay prior to enrollment (days) 2.5 (3.6)
No. (%) with:
Respiratory disease 6 (19)
Hepatic dysfunction 4 (13)
Malignancy 2 (6)
Diabetes mellitus 13 (42)
Chronic kidney disease 8 (26)
Mean (SD) APACHE II score on enrollmenta 21 (9.0)
Mean (SD) ICU length of stay after enrollment (days) 17 (27)
No. (%) of in-hospital mortality 5 (16)
a
Scores were unable to be calculated for 12 patients with bacterial sepsis and 6 patients with other causes of SIRS due to missing arterial blood gases.
The goals of the identified biomarker combinations were to maximize
the NPV, given that the highest utility of the derived algorithm in the
MICU population would be to identify patients at low likelihood of bac-
terial infection in whom empirical antibiotics could be safely discontin-
ued, and to maximize the specificity, in order to identify the combination
of biomarkers that would characterize a significant number of patients
without bacterial infection as test negative.
For all calculations, a 2-tailed P value of ⬍0.05 was considered signif-
icant. All statistical calculations were performed using commercially
available software: SAS version 9.3 (SAS Institute, Inc., Cary, NC) and
STATA version 13.0 (StataCorp LP, College Station, TX).
RESULTS
A total of 286 patients were screened for eligibility, with 70 (24%)
patients fulfilling all study inclusion criteria. Reasons for exclu-
sion included the following: lack of SIRS criteria (n ⫽ 5); antibi-
otics received for ⬎4 h past the baseline biomarker measurement
time point (n ⫽ 5); cardiopulmonary arrest (n ⫽ 5); immunosup-
pression (n ⫽ 147); current receipt of broad-spectrum antibiotic
therapy for a recently documented bacterial infection (n ⫽ 25); do
not resuscitate status (n ⫽ 7); and the inability to obtain residual
blood samples for all time-point measurements (n ⫽ 22).
A total of 31 (44%) patients had bacterial infection, as de-
termined via medical record review (with 96% agreement be-
tween the two investigators on initial independent review). The
source of infection was determined as blood (n ⫽ 14; 45%),
pulmonary (n ⫽ 10; 32%), urinary (n ⫽ 3; 10%), spontaneous
bacterial peritonitis (n ⫽ 2; 7%), mastoiditis (n ⫽ 1; 3%), and
Clostridium difficile infection (n ⫽ 1; 3%). Etiologies of other
causes of SIRS included conditions such as hepatic decompen-
sation, diabetic ketoacidosis, alcohol withdrawal, pancreatitis,
infections due to Candida spp., and cardiogenic shock. Charac-
teristics of the study population with and without bacterial sepsis
are shown in Table 2. Patients with bacterial sepsis were less likely
than those with other causes of SIRS to have hepatic dysfunction
(13% versus 44%, respectively; P ⫽ 0.008) or malignancy (6%
versus 33%, respectively; P ⫽ 0.008). The mean APACHE II score
on study enrollment for patients with bacterial sepsis versus for
those with other causes of SIRS was 22 (standard deviation [SD],
9.0) versus 24 (SD, 9.0), respectively. There were no significant
differences between groups in all-cause hospital mortality (16%
6496 aac.asm.org Antimicrobial Agents and Chemotherapy
Other causes of SIRS (n ⫽ 39) P
54.4 (16.6) 0.14
23 (59) 0.81
16 (41) 0.21
2.2 (2.9) 0.15
3.8 (4.6) 0.30
4 (10) 0.32
17 (44) 0.008
13 (33) 0.008
10 (26) 0.20
13 (33) 0.60
Downloaded from http://aac.asm.org/ on December 11, 2019 by guest
23 (9.0) 0.61
6.7 (6.9) 0.08
10 (26) 0.39
versus 26% for bacterial sepsis versus other causes of SIRS, respec-
tively; P ⫽ 0.39), ICU mortality (P ⫽ 0.53), or 14-day mortality
(P ⫽ 0.28).
The mean (SD) times elapsed from the baseline biomarker
measurement to the follow-up time-point measurements were as
follows: 24 h (mean, 24 h; SD, 5.2 h), 48 h (mean, 48 h; SD, 6.4 h),
and 72 h (mean, 71 h; SD, 7.4 h). Mean (SD) values of the nine
candidate biomarkers for patients with bacterial sepsis versus for
those with other causes of SIRS are shown in Table 3. While mean
values of haptoglobin at the 48-h time point and serum amyloid P
at baseline were significantly different between the two groups of
patients, only CRP and PCT demonstrated significantly higher
mean values for patients with bacterial sepsis than for those with
other causes of SIRS at all of the measured time points (Fig. 1; see
Fig. S1 in the supplemental material for the associated scatter
plots). Therefore, CRP and PCT were selected as the most prom-
ising biomarkers for differentiating between bacterial sepsis and
other causes of SIRS in the study population, and NPVs and spec-
ificities, along with 95% confidence intervals (CIs) for the combi-
nations of PCT and CRP at differing time points and cutoff values,
were calculated. Table 4 summarizes the combinations of time
points and cutoff values for PCT and CRP that maximized NPVs
and specificities.
Figure 2 depicts ROC curves for the individual biomarkers as
well as for the combination of PCT and CRP (and their interac-
tion), all at the 24-h time point. The combination of PCT and CRP
at the 24-h time point demonstrated a higher AUC (0.810) than
the 24-h PCT and CRP individually (AUCs of 0.782 and 0.759,
respectively).
DISCUSSION
We conducted a prospective cohort study to evaluate the utility of
nine biomarkers in differentiating bacterial sepsis from other
causes of SIRS in the MICU patient population. The results of our
study support the use of a combination of PCT and CRP, with the
highest achievable NPV of ⬃86% and specificity of ⬃80%, de-
pending on the specific time points and cutoff values utilized.
SIRS is common in the MICU population, and, given the asso-
ciation with reduction in mortality (1, 2), patients with SIRS will
often receive early empirical broad-spectrum antibiotic therapy.
October 2015 Volume 59 Number 10
 Biomarkers and Sepsis Diagnosis

TABLE 3 Biomarkers among medical intensive care unit patients with of antibiotics are often used to treat MICU patients with a non-
bacterial sepsis and other causes of SIRS bacterial etiology of SIRS, contributing to overuse of antibiotics in
Mean value (SD) with: this setting. Thus, there is a need for reliable diagnostic biomark-
ers as an adjunct in the overall clinical decision-making process
Bacterial sepsis Other causes of
Biomarker (n ⫽ 31) SIRS (n ⫽ 39) P
for differentiating bacterial sepsis from other causes of SIRS.
Given that the greatest clinical utility for biomarkers in diag-
␣-2 Macroglobulin (mg/dl)
nosing bacterial sepsis specifically in the MICU setting would
Baseline 126 (52.1) 115 (41.8) 0.41
24 h 101 (43.0) 102 (34.6) 0.66
likely be to facilitate the discontinuation of empirical antibiotics in
48 h 101 (43.8) 101 (46.4) 0.90 patients without bacterial infection, the goal of our study was to
72 h 102 (42.0) 104 (48.8) 0.89 derive a combination of biomarkers with a high NPV. In addition,
a high specificity was important, so that a significant number of
C-reactive protein (mg/liter) patients without bacterial infection would test negative using the
Baseline 113 (98.2) 52.0 (56.9) 0.01 algorithm’s cutoffs. Previous studies assessing the use of biomark-
⬍0.001

Downloaded from http://aac.asm.org/ on December 11, 2019 by guest

24 h 134 (64.5) 73.2 (65.8) ers for the diagnosis of bacterial sepsis have demonstrated variable
48 h 123 (75.3) 59.9 (49.0) ⬍0.001
performance characteristics (5–15), including NPV values rang-
72 h 99.9 (72.4) 51.3 (43.7) 0.004
ing from 35% to 97% and specificity values ranging from 55% to
Ferritin (ng/ml) 91%. The heterogeneity in these results, which has been noted in a
Baseline 210 (609) 331 (610) 0.05 recent meta-analysis (24), is likely a result of differences in patient
24 h 229 (374) 304 (453) 0.47 populations (e.g., inclusion of surgical ICU patients), the type of
48 h 361 (824) 396 (705) 0.48 assay performed, the timing of sample testing, the cutoff values
72 h 219 (550) 483 (850) 0.11 selected, and selection of the gold standard for infection. In addi-
Fibrinogen (␮g/ml)
tion, despite the fact that sepsis is a complex process associated
Baseline 56.6 (278) 12.3 (34.8) 0.63 with the production of a number of inflammatory biomarkers, the
24 h 6.76 (3.91) 23.3 (71.3) 0.96 majority of studies have focused solely on the utilization of PCT
48 h 6.62 (3.63) 7.42 (7.13) 0.85 (5–9, 11, 12, 14, 15).
72 h 5.83 (3.07) 7.72 (6.13) 0.33

Haptoglobin (mg/dl)
Baseline 83.0 (30.2) 69.6 (59.3) 0.07
24 h 87.1 (47.4) 73.7 (78.7) 0.09
48 h 93.5 (47.0) 73.3 (79.1) 0.02
72 h 96.9 (59.4) 77.9 (75.6) 0.07

Procalcitonin (ng/ml)
Baseline 23.1 (49.9) 2.97 (8.11) 0.01
24 h 26.1 (49.3) 3.60 (8.24) ⬍0.001
48 h 22.3 (42.0) 5.26 (15.3) ⬍0.001
72 h 11.5 (19.7) 4.52 (15.1) 0.009

Serum amyloid A (␮g/ml)

Baseline 13.8 (16.9) 37.8 (27.9) 0.08
24 h 10.7 (4.25) 10.7 (11.7) 0.75
48 h 14.7 (13.8) 10.0 (5.70) 0.36
72 h 9.93 (4.10) 10.1 (7.70) 0.80

Serum amyloid P (mg/liter)

Baseline 490 (36) 331 (45) 0.02
24 h 37.9 (15.0) 35.2 (21.7) 0.22
48 h 41.7 (17.6) 34.3 (18.9) 0.05
72 h 43.1 (18.7) 35.2 (18.1) 0.09

Tissue plasminogen activator

(ng/ml)
Baseline 7.82 (5.30) 8.06 (5.65) 0.95
24 h 5.63 (2.76) 6.81 (5.15) 0.79
48 h 5.52 (3.36) 6.98 (4.92) 0.37
72 h 4.90 (3.26) 7.81 (8.46) 0.06

However, patients with bacterial sepsis can present with nonspe-

cific signs (e.g., fever, tachycardia), and it can be difficult to dif-
ferentiate bacterial sepsis from other causes of SIRS (i.e., nonbac- FIG 1 Trend of PCT and CRP concentrations during follow-up time point
terial sepsis or noninfectious SIRS). As a result, prolonged courses measurements in patients with bacterial sepsis versus other causes of SIRS.

October 2015 Volume 59 Number 10 Antimicrobial Agents and Chemotherapy aac.asm.org 6497
Han et al.

TABLE 4 Performance characteristics of combinations of PCT and CRP in diagnosing bacterial sepsis
PCT (ng/ml) and CRP (mg/liter) combination NPV (95% CI) No. of test negatives Specificity (95% CI)
24-h PCT ⱖ0.5 and 24-h CRP ⱖ40 85.7 (72.7–98.7) 24 63.2 (47.9–78.5)
48-h PCT ⱖ0.5 and 48-h CRP ⱖ50 83.3 (70.0–96.6) 25 64.1 (49.0–79.2)
48-h PCT ⱖ1.0 and 24-h CRP ⱖ40 82.4 (69.6–95.2) 28 71.8 (57.7–85.9)
24-h PCT ⱖ1.5 and 24-h CRP ⱖ40 81.1 (68.5–93.7) 30 78.9 (68.5–65.9)
48-h PCT ⱖ2.0 and 48-h CRP ⱖ40 81.1 (68.5–93.7) 30 76.9 (63.7–90.1)
48-h PCT ⱖ2.0 and 24-h CRP ⱖ40 81.6 (69.3–93.9) 31 79.5 (66.8–92.2)

In the present study, we focused specifically on medical ICU cohort. For example, a PCT of ⱖ0.5 ng/ml and a CRP of ⱖ40
patients and determined biomarker values prospectively at multi- mg/liter, both measured at 24 h after the onset of suspected bac-
ple consecutive time points. We identified a number of combina- terial sepsis, demonstrated an NPV of ⬃86% in our study and

Downloaded from http://aac.asm.org/ on December 11, 2019 by guest

tions of PCT and CRP cutoff values and measurement time points designated ⬃63% of patients without bacterial infection as testing
for differentiating bacterial sepsis from other causes of SIRS in this negative using this combination of biomarkers. Similarly, a PCT

FIG 2 Receiver-operating characteristic (ROC) curves at the 24-h time point for PCT (a), CRP (b), combination of PCT and CRP (c), and combination of PCT
and CRP with interaction (d).

6498 aac.asm.org Antimicrobial Agents and Chemotherapy October 2015 Volume 59 Number 10

of ⱖ2.0 ng/ml and a CRP of ⱖ40 mg/liter measured at the 48-h
time point demonstrated an NPV of ⬃81% and a specificity of
⬃77%.
Notably, combinations of PCT and CRP demonstrated more
favorable combined NPV and specificity values than the individ-
ual biomarkers alone. For example, using a cutoff value for PCT
of ⱖ1.5 ng/ml at the 24-h time point demonstrated an NPV of
⬃78% and a specificity of 66%. Using a cutoff value for CRP
of ⱖ50 mg/liter at the 24-h time point demonstrated an NPV of
⬃83% and a specificity of ⬃51%. Similarly, the discriminatory
ability of the combination of PCT and CRP was excellent (AUC of
0.810 for the 24-h time point), and superior to that of the individ-
ual biomarkers.
These results have important implications for the accurate di-
agnosis of bacterial sepsis in the MICU. PCT and CRP are widely
used biomarkers in the clinical laboratory, with rapid turnaround
times that facilitate their early use in the course of clinical care.
The high NPVs and specificities determined in our study for var-
ious cutoff values and measurement time points suggest that a
combination of these biomarkers could be of significant utility in
curtailing antibiotic use in critically ill patients who are receiving
broad-spectrum antibiotics with nonbacterial etiologies of SIRS.
The specific combination of PCT and CRP cutoff values and mea-
surement time points selected would ultimately depend on the
relative importance placed on NPV versus the ability to identify a
greater number of patients without true bacterial infection. In
addition, given these performance characteristics, future studies
should focus on the prognostic utility of combinations of PCT and
CRP for predicting mortality in the MICU setting.
There are potential limitations of this study. There may have
been misclassification of the determination of bacterial infection.
However, there was 96% agreement between the two investiga-
tors, who were blinded to the biomarker results, on the initial
independent review. The small numbers of patients in this study
was a limitation, and this may have precluded our ability to detect
certain significant associations. We did not have a large enough
number of patients to determine optimal biomarker combina-
tions for specific sources of infections (e.g., bacteremia). The der-
ivation cohort specifically excluded significantly immunocom-
promised patients, because it is unlikely that empirical antibiotics
would be discontinued in this population (regardless of bio-
marker results), decreasing the utility of such an algorithm in this
group. Finally, performance characteristics of biomarker combi-
nations were derived from a cohort of patients in a single academic
medical center and, thus, may not be generalizable to dissimilar
settings.
In conclusion, combinations of PCT and CRP demonstrated a
high discriminative ability for bacterial sepsis versus other causes
of SIRS in our cohort of MICU patients. These biomarkers, there-
fore, may be of significant utility for diagnosing bacterial sepsis in
this population when interpreted in the context of other clinical
and laboratory assessments. Future studies should focus on the
development of algorithms using a combination of biomarkers to
reduce unnecessary antibiotic use in the ICU setting.
ACKNOWLEDGMENTS
This work was supported by the CDC cooperative agreement FOA CK11-
001—Epicenters for the Prevention of Healthcare-Associated Infections
(to E.L.) and the National Institutes of Health grant K01-AI103028 (to
J.H.H.).
October 2015 Volume 59 Number 10 Antimicrobial Agents and Chemotherapy
Biomarkers and Sepsis Diagnosis
The funding agencies had no role in the design and conduct of the
study, collection, management, analysis, and interpretation of the data, or
preparation, review, or approval of the manuscript.
We report no potential conflicts of interest.
REFERENCES
1. Carlet J, Ben Ali A, Chalfine A. 2004. Epidemiology and control of
antibiotic resistance in the intensive care unit. Curr Opin Infect Dis 17:
309 –316. http://dx.doi.org/10.1097/01.qco.0000136927.29802.68.
2. Zahar JR, Timsit JF, Garrouste-Orgeas M, Francais A, Vesin A, De-
scorps-Declere A, Dubois Y, Souweine B, Haouache H, Goldgran-
Toledano D, Allaouchiche B, Azoulay E, Adrie C. 2011. Outcomes in
severe sepsis and patients with septic shock: pathogen species and infec-
tion sites are not associated with mortality. Crit Care Med 39:1886 –1895.
http://dx.doi.org/10.1097/CCM.0b013e31821b827c.
Downloaded from http://aac.asm.org/ on December 11, 2019 by guest
3. Rice LB. 2009. The clinical consequences of antimicrobial resistance. Curr
Opin Microbiol 12:476 – 481. http://dx.doi.org/10.1016/j.mib.2009.08
.001.
4. Paterson DL. 2004. Collateral damage from cephalosporin or quinolone
antibiotic therapy. Clin Infect Dis 38(Suppl 4):S341–S345. http://dx.doi
.org/10.1086/382690.
5. Harbarth S, Holeckova K, Froidevaux C, Pittet D, Ricou B, Grau GE,
Vadas L, Pugin J, Geneva Sepsis N. 2001. Diagnostic value of procalci-
tonin, interleukin 6, and interleukin 8 in critically ill patients admitted
with suspected sepsis. Am J Respir Crit Care Med 164:396 – 402. http://dx
.doi.org/10.1164/ajrccm.164.3.2009052.
6. Balcl C, Sungurtekin H, Gurses E, Sungurtekin U, Kaptanoglu B. 2003.
Usefulness of procalcitonin for diagnosis of sepsis in the intensive care
unit. Crit Care 7:85–90. http://dx.doi.org/10.1186/cc1843.
7. Garnacho-Montero J, Huici-Moreno MJ, Gutierrez-Pizarraya A, Lopez
I, Marquez-Vacaro JA, Macher H, Guerrero JM, Puppo-Moreno A.
2014. Prognostic and diagnostic value of eosinopenia, C-reactive protein,
procalcitonin, and circulating cell-free DNA in critically ill patients admit-
ted with suspicion of sepsis. Crit Care 18:R116. http://dx.doi.org/10.1186
/cc13908.
8. Meynaar IA, Droog W, Batstra M, Vreede R, Herbrink P. 2011. In
critically ill patients, serum procalcitonin is more useful in differentiating
between sepsis and SIRS than CRP, Il-6, or LBP. Crit Care Res Pract 2011:
594645. http://dx.doi.org/10.1155/2011/594645.
9. Ruiz-Alvarez MJ, Garcia-Valdecasas S, De Pablo R, Sanchez Garcia M,
Coca C, Groeneveld TW, Roos A, Daha MR, Arribas I. 2009. Diagnostic
efficacy and prognostic value of serum procalcitonin concentration in
patients with suspected sepsis. J Intensive Care Med 24:63–71. http://dx
.doi.org/10.1177/0885066608327095.
10. Selberg O, Hecker H, Martin M, Klos A, Bautsch W, Kohl J. 2000.
Discrimination of sepsis and systemic inflammatory response syndrome
by determination of circulating plasma concentrations of procalcitonin,
protein complement 3a, and interleukin-6. Crit Care Med 28:2793–2798.
http://dx.doi.org/10.1097/00003246-200008000-00019.
11. Suprin E, Camus C, Gacouin A, Le Tulzo Y, Lavoue S, Feuillu A,
Thomas R. 2000. Procalcitonin: a valuable indicator of infection in a
medical ICU? Intensive Care Med 26:1232–1238. http://dx.doi.org/10
.1007/s001340000580.
12. Castelli GP, Pognani C, Meisner M, Stuani A, Bellomi D, Sgarbi L.
2004. Procalcitonin and C-reactive protein during systemic inflammatory
response syndrome, sepsis and organ dysfunction. Crit Care 8:R234 –
R242. http://dx.doi.org/10.1186/cc2877.
13. Ugarte H, Silva E, Mercan D, De Mendonca A, Vincent JL. 1999.
Procalcitonin used as a marker of infection in the intensive care unit. Crit
Care Med 27:498 –504. http://dx.doi.org/10.1097/00003246-199903000
-00024.
14. Tsangaris I, Plachouras D, Kavatha D, Gourgoulis GM, Tsantes A,
Kopterides P, Tsaknis G, Dimopoulou I, Orfanos S, Giamarellos-
Bourboulis E, Giamarellou H, Armaganidis A. 2009. Diagnostic and
prognostic value of procalcitonin among febrile critically ill patients with
prolonged ICU stay. BMC Infect Dis 9:213. http://dx.doi.org/10.1186
/1471-2334-9-213.
15. Brunkhorst FM, Wegscheider K, Forycki ZF, Brunkhorst R. 2000.
Procalcitonin for early diagnosis and differentiation of SIRS, sepsis, severe
sepsis, and septic shock. Intensive Care Med 26(Suppl 2):S148 –S152. http:
//dx.doi.org/10.1007/BF02900728.
16. Nobre V, Harbarth S, Graf JD, Rohner P, Pugin J. 2008. Use of procal-
aac.asm.org 6499
Han et al.
citonin to shorten antibiotic treatment duration in septic patients: a ran-
domized trial. Am J Respir Crit Care Med 177:498 –505. http://dx.doi.org
/10.1164/rccm.200708-1238OC.
17. Aouifi A, Piriou V, Bastien O, Blanc P, Bouvier H, Evans R, Celard M,
Vandenesch F, Rousson R, Lehot JJ. 2000. Usefulness of procalcitonin
for diagnosis of infection in cardiac surgical patients. Crit Care Med 28:
3171–3176. http://dx.doi.org/10.1097/00003246-200009000-00008.
18. Christ-Crain M, Stolz D, Bingisser R, Muller C, Miedinger D, Huber
PR, Zimmerli W, Harbarth S, Tamm M, Muller B. 2006. Procalcitonin
guidance of antibiotic therapy in community-acquired pneumonia: a ran-
domized trial. Am J Respir Crit Care Med 174:84 –93. http://dx.doi.org/10
.1164/rccm.200512-1922OC.
19. Enguix A, Rey C, Concha A, Medina A, Coto D, Dieguez MA. 2001.
Comparison of procalcitonin with C-reactive protein and serum amyloid
for the early diagnosis of bacterial sepsis in critically ill neonates and
children. Intensive Care Med 27:211–215. http://dx.doi.org/10.1007
/s001340000709.
20. de Kruif MD, Limper M, Gerritsen H, Spek CA, Brandjes DP, ten Cate
H, Bossuyt PM, Reitsma PH, van Gorp EC. 2010. Additional value of
procalcitonin for diagnosis of infection in patients with fever at the emer-
gency department. Crit Care Med 38:457– 463. http://dx.doi.org/10.1097
/CCM.0b013e3181b9ec33.
21. Briel M, Schuetz P, Mueller B, Young J, Schild U, Nusbaumer C, Periat
P, Bucher HC, Christ-Crain M. 2008. Procalcitonin-guided antibiotic
use versus a standard approach for acute respiratory tract infections in
primary care. Arch Intern Med 168:2000 –2007 (discussion, 2007–2008).
http://dx.doi.org/10.1001/archinte.168.18.2000.
22. Uzzan B, Cohen R, Nicolas P, Cucherat M, Perret GY. 2006. Procalci-
tonin as a diagnostic test for sepsis in critically ill adults and after surgery
or trauma: a systematic review and meta-analysis. Crit Care Med 34:1996 –
2003. http://dx.doi.org/10.1097/01.CCM.0000226413.54364.36.
6500 aac.asm.org Antimicrobial Agents and Chemotherapy
23. Tang BM, Eslick GD, Craig JC, McLean AS. 2007. Accuracy of procal-
citonin for sepsis diagnosis in critically ill patients: systematic review and
meta-analysis. Lancet Infect Dis 7:210 –217. http://dx.doi.org/10.1016
/S1473-3099(07)70052-X.
24. Wacker C, Prkno A, Brunkhorst FM, Schlattmann P. 2013. Procalci-
tonin as a diagnostic marker for sepsis: a systematic review and meta-
analysis. Lancet Infect Dis 13:426 – 435. http://dx.doi.org/10.1016/S1473
-3099(12)70323-7.
25. Pierrakos C, Vincent JL. 2010. Sepsis biomarkers: a review. Crit Care
14:R15. http://dx.doi.org/10.1186/cc8872.
26. Levy MM, Fink MP, Marshall JC, Abraham E, Angus D, Cook D, Cohen
J, Opal SM, Vincent JL, Ramsay G, SCCM/ESICM/ACCP/ATA/SIS.
2003. 2001 SCCM/ESICM/ACCP/ATS/SIS International Sepsis Defini-
tions Conference. Crit Care Med 31:1250 –1256. http://dx.doi.org/10
.1097/01.CCM.0000050454.01978.3B.
27. Bone RC, Balk RA, Cerra FB, Dellinger RP, Fein AM, Knaus WA,
Downloaded from http://aac.asm.org/ on December 11, 2019 by guest
Schein RM, Sibbald WJ, ACCP/SCCM Consensus Conference Commit-
tee. 2009. Definitions for sepsis and organ failure and guidelines for the
use of innovative therapies in sepsis. The ACCP/SCCM Consensus Con-
ference Committee. American College of Chest Physicians/Society of Crit-
ical Care Medicine. 1992. Chest 136:e28. http://dx.doi.org/10.1378/chest
.101.6.1644.
28. Meisner M. 2014. Update on procalcitonin measurements. Ann Lab Med
34:263–273. http://dx.doi.org/10.3343/alm.2014.34.4.263.
29. Knaus WA, Drapier EA, Wagner DP, Zimmerman JE. 1985. APACHE
II: a severity of disease classification system. Crit Care Med 13:818 – 829.
http://dx.doi.org/10.1097/00003246-198510000-00009.
30. Horan TC, Andrus M, Dudeck MA. 2008. CDC/NHSN surveillance
definition of health care-associated infection and criteria for specific types
of infections in the acute care setting. Am J Infect Control 36:309 –332.
http://dx.doi.org/10.1016/j.ajic.2008.03.002.
October 2015 Volume 59 Number 10