The role of procalcitonin as a predictor of nosocomial sepsis inpreterm infants

DAN TURNER1, CATHY HAMMERMAN2, BERNARD RUDENSKY3,

YECHIEL SCHLESINGER4 & MICHAEL S. SCHIMMEL2

Departments of 1Pediatrics and 2Neonatology, 3Clinical Microbiology Laboratory, and 4Pediatric Infectious Disease Unit,

Shaare Zedek Medical Center, Jerusalem, Israel

AbstractAim: To assess the role of procalcitonin in detecting nosocomial sepsis in preterm infants, after the onset of clinicalsymptoms. Subjects: 100 preterm infants, 24�36 wk of gestation, were followed from the age of 3 d until discharge.Procalcitonin and C-reactive protein (CRP) levels were measured within 3 d of sepsis workup events. Results: 141 bloodsamples were drawn from 36 infants during 85 episodes of sepsis workup performed between 4 and 66 d of life. Of theseepisodes, 51 (60%) were not a result of documented sepsis and thereby served as the negative comparison group. Medianprocalcitonin levels were higher in the septic group compared with the non-septic group at the time of the sepsis workup(2.7 vs 0.5 ng/ml, p�/0.003), at 1�24 h after the sepsis workup (4.6 vs 0.6 ng/ml, p�/0.003), and at 25�48 h (6.9 vs 2.0 ng/ml, p�/0.016). Using high cutoff levels, both procalcitonin (2.3 ng/ml) and CRP (30 mg/l) had high specificity and positivepredictive value (97%, 91% and 96%, 87%, respectively) but low sensitivity (48% and 41%, respectively) to detect sepsis.Areas under the ROC curve for procalcitonin and CRP were 0.74 and 0.73, respectively.

Conclusion: Procalcitonin �/2.3 ng/ml or CRP �/30 mg/l indicates a high likelihood for neonatal sepsis, and antibiotictherapy should be continued even in the presence of sterile cultures.

Key Words: Diagnosis, late-onset sepsis, nosocomial sepsis, procalcitonin, preterm infants

Introduction

Bacterial infections are a leading cause of morbidity

and mortality in preterm infants [1], but the clinical

signs are often non-specific, even in serious invasive

disease. Thus, the decision to start antibiotic

therapy is frequently based on non-specific clinical

clues, and numerous infants are treated with pro-

longed courses of antibiotics in the absence of

proven infection. Procalcitonin (PCT) has been

reported to be an excellent diagnostic test for

detection of invasive bacterial infections in adults

and children [2�4]. PCT increases in serum within

2�3 h of infection onset, peaks by 6�12 h, and

reverts to normal within 2 d [4,5]. PCT is also

useful in neonatal sepsis, but the evidence is less

convincing, especially in preterm infants [3]. Pre-

vious studies have assessed PCT in preterm infants

with early-onset sepsis [3,6], but only two [7,8]

evaluated nosocomial infections in preterm infants,

with conflicting results. We followed a cohort of

preterm infants to assess the ability of PCT to

detect nosocomial sepsis.

Methods

Patients

The study was performed in a 30-bed neonatal

intensive care unit (NICU) at Shaare Zedek Medical

Center, Israel. The hospital serves approximately

10 000 births per year, with 7.8% weighing less than

2500 g. Preterm infants (36 wk or less) admitted to

the NICU between September 2003 and January

2004 were eligible for enrollment. Informed parental

consent was obtained according to the local Institu-

tional Review Board’s instructions.

Sample collection

Since PCT concentration is known to increase during

the first 3d of life even in healthy new-borns, follow up

(Received 11 November 2005; revised 2 March 2006; accepted 21 April 2006)

ISSN 0803-5253 print/ISSN 1651-2227 online # 2006 Taylor & Francis

DOI: 10.1080/08035250600767811

Correspondence: Dan Turner, Division of Pediatric Gastroenterology, Hepatology, and Nutrition, Hospital for Sick Children, 555 University Ave, Toronto,

M5G 1X8, Canada. Tel: �/1 416 813 6555. Fax: �/1 416 813 6531. E-mail: dan.turner@sickkids.ca

Acta Pædiatrica, 2006; 95: 1571�1576

was initiated on day-3 of age and continued daily until

discharge or death [9,10]. PCT was measured as part

of routine blood tests, during each clinical event

suspicious of sepsis: at the time of the sepsis workup

(SWU) and 1�24, 25�48, and 49�72 h thereafter.

The samples obtained at the time of the SWU were

also tested for C-reactive protein (CRP) level. Infants

were classified into four groups (Table I): group 1:

culture-proven sepsis; group 2: clinical sepsis; group

3: uncertain; and group 4: uninfected [9,10]. The

classification, which was computed at the sampling

time of every PCT and CRP value, was stringent so as

to minimize measurement bias. Investigators who

classified the groups and physicians who managed

the infants were all blinded to the levels of PCT and

CRP.

Definitions

Recurrent episodes of sepsis were considered to be

separate if the infant was clinically stable during an

interval of at least 1 wk without antibiotic therapy and

negative repeated cultures. Coagulase-negative Sta-

phylococcus was considered pathogenic if it grew in

two separate cultures of normally sterile body fluids.

Urine was collected using the supra-pubic aspiration

technique. Pneumonia was diagnosed on the basis of

pulmonary symptoms with acute radiological find-

ings.

PCT and CRP assays

Samples of 0.5 ml of blood were drawn into ethyle-

nediaminetetraacetic acid. Samples were separated,

and the plasma frozen at �/70oC. At the end of the

study period, the plasma samples were thawed, and

analyzed for PCT concentrations by an immunolu-

minometric assay (LUMItest# PCT, BRAHMS

Diagnostica, Berlin, Germany), as previously de-

scribed [9]. The manufacturer claims an analytical

assay sensitivity of 0.1 mg/l and a functional assay

sensitivity of 0.3 mg/l. The same plasma samples were

then measured for CRP concentration by immuno-

nephelometric assay using the Behring Nephelometry

Analyzer (BN2, Behring, Marburg, Germany). This

method has a limit of detection of 0.02 mg/l and a

limit of quantification of 0.15 mg/l [13].

Statistical analyses

Data are presented as means (9/standard deviation)

or medians (interquartile range) and compared

using the unpaired Student’s t-test, Mann-Whitney

rank-sum test or the Kruskal-Wallis on ranks, as

appropriate for the distribution normality. Correla-

tions between individual parameters were sought

using a Spearman correlation. To determine the

relative contributions of explanatory variables to

PCT levels, multivariate regression analysis was

used. Area under the receiver operating character-

istic (ROC) curve (9/95% CI) of over 0.7 was

considered as a fair diagnostic test, and over 0.9 as

a very good test. Interval likelihood ratios were

computed by dividing the proportion of septic

infants in the value interval with the proportion of

the non-septic infants in the same interval. All

comparisons were made using two-sided significance

levels of p B/0.05. Statistical analysis was performed

using SPSS v. 12.0.

Table I. Definition criteria of the four sepsis groups.

Clinical signs of infection (positive if]/two of the following)

Respiratory Any of: tachypnea/hypopnea/apnea/cyanosis

Cardiovascular Any of: bradycardia/tachycardia

Shock Any of: poor perfusion/low blood pressure

Consciousness Any of: irritability/lethargy/poor feeding/hypotonia/seizures

Others Any of: hepatosplenomegaly/jaundice/fever

Sepsis screen (positive if]/two of the following)

WBC Any of: leukocytosis/leukocytopenia

ANC Any of: neutrophilia/neutropenia

Thrombocytes Any of: thrombocytopenia/thrombocytosis

CSF Any of: pleocytosis/Gram-positive stain

Glucose Any of: 40 mg%�/glucose�/180 mg%

Sepsis groups

1) Confirmed sepsis Positive blood, CSF or urine culture/radiologically proven pneumonia/cellulitis and clinical signs of infection

2) Clinical sepsis Clinical signs of infection and positive sepsis screen

3) Uncertain No fulfillment of the above criteria and no change in status after 3 d, requiring longer antibiotic therapy

4) No infection No fulfillment of the above criteria and clinical condition improved within 3 d

Reference values for laboratory results were as described in detail elsewhere [11,12]. Poor perfusion was defined as corpus capillary refill

�/3 s.

WBC: white blood cells; ANC: absolute neutrophil count; CSF: cerebrospinal fluid.

1572 D. Turner et al.

Results

Demographic and sample characteristics

Parents of 100 of 110 preterm infants admitted to the

NICU during the study period consented to be

followed in the study. A total of 141 blood samples

were obtained for PCT levels from 36 infants during

85 episodes of SWU performed between 4 and 66 d of

life (mean 18.19/13.1 d). Gestational age and birth-

weight of the 36 infants were 329/2.9 wk (range 24�36 wk) and 16829/500 g (range 479�2615 g),

respectively.

The 85 SWU episodes were classified as defined in

Table I: 28 (33%) in group 1, six (7%) in group 2, 18

(21%) in group 3, and 33 (39%) in group 4. Twenty

pathogens were isolated from the infants of group 1,

including four coagulase-negative and one coagulase-

positive Staphylococci , four Escherichia coli , three

Klebsiella pneumoniea , two Serratia , two Enterobacter

cloaca , two Pseudomonas aeruginosa , one Candida

albicans , and one Morganella morganii .

No differences in PCT and CRP levels were found

between groups 1 and 2 or between groups 3 and 4 at

any time after the SWU (Kruskal-Wallis, all p �/0.05).

Therefore, the two septic subgroups (1 and 2) and the

two non-septic subgroups (3 and 4) were combined

for subsequent statistical analyses. Infants were

sampled twice in 66% of the episodes and once in

the rest, at one or two of the four time windows used

for statistical comparisons. In any case, no more than

one sample per infant was allowed at a specific time

interval, and thus no measurement from a given infant

was repeated in the analysis. There was no difference

in the number of samples per infant obtained from the

septic group (1.72 samples per episode) versus the

control group (1.62 samples per episode; p�/0.8).

The 51 episodes of SWU in groups 3 and 4

combined served as the comparison to the study

group (groups 1 and 2). No differences were found

(Student’s t-test) between the study and the compar-

ison groups in birthweight (11449/504 vs 13829/499

g, p �/0.05) and gestational age (28.89/3.4 vs 30.19/3

wk, p �/0.05). Similarly, there were also no differences

in the mean weight and corrected gestational age at

the time of the SWU (15299/678 vs 16109/629 g, p�/

0.92; and 31.49/4.1 vs 31.59/3.1 wk, p�/0.59).

Moreover, PCT and CRP levels did not correlate

with either birthweight or gestational age (actual and

corrected) at any time interval after the SWU (Spear-

man correlation, �/0.3B/r B/0.14, all p �/0.05), and

thus no group sub-analysis was performed. Four

infants died, one from the non-septic and three from

the septic groups.

Group comparisons

Median PCT values (interquartile range) of the septic

groups were significantly higher than those of the

comparison group (Figure 1): at the time of the SWU

(2.7 (0.8�5.4) vs 0.5 (0.46�0.63) ng/ml; p�/0.003,

Mann-Whitney test), at 1�24 h (4.6 (2.5�14.9) vs 0.6

(0.5�1.3) ng/ml; p�/0.003), and at 25�48 h (6.9

(4.7�16.3) vs 2 (0.5�3.5) ng/ml; p�/0.016). Similarly

to PCT, median CRP values of the septic groups were

also significantly higher than those of the comparison

groups (19 (10�80) vs 12 (8�15) mg/l at the time of

the SWU; p�/0.001, Mann-Whitney test).

PCT and CRP utilization

A ROC curve was plotted to compare the usefulness

of PCT and CRP to detect sepsis at the time of the

SWU (Figure 2). Area under the curve for PCT and

CRP was 0.74 (95% CI 0.59�0.88) and 0.73 (0.59�0.84), respectively, indicating that the tests are

comparable in discriminating septic from non-septic

infants. The best cutoffs of PCTand CRP values were

0.74 ng/ml and 1.4 mg/l, resulting in a sensitivity/

01

23

45

67

8

h 27-94h 84-52h 42-1h 0

spuorg citpeS

citpes-noNspuorg

32=n

93=n

51=n

81=n

01=n

9=n12=n

6=n

eM

di a

nP

CT

l eve

l

pukrow sispes retfa emiT

p 300.0=

p 610.0=

p 300.0=

SN

Figure 1. Median PCT values (ng/ml) at the time of sepsis workup and 3 d thereafter, according to septic (groups 1 and 2) and non-septic

(groups 3 and 4) infants. There was no difference in the mean sampling time between the septic and the non-septic infants at each of the

three time intervals (Student’s t -test, all p �/0.05). NS: not significant.

Procalcitonin in preterm infants 1573

specificity of 70%/67% and 72%/71%, respectively.

Areas under the curve for PCT at 1�24, 25�48, and

49�72 h after the SWU were 0.84 (0.67�1.00), 0.79

(0.56�1.00), and 0.67 (0.4�0.95), respectively.

Higher PCT and CRP cutoffs values were then

evaluated to obtain higher specificity, as shown in

Table II. Likelihood ratio (LR) represents how many

times it is more (or less) likely that a PCT result

will be found in septic compared with non-septic

infants [14]. PCT levels of over 3 ng/ml and CRP

over 30 mg/l showed very high LR to detect sepsis

(Table III).

To test whether PCT or CRP values could predict

response to treatment, the responsible physician

completed, after 72 h of treatment, a global assess-

ment of whether the clinical condition improved,

deteriorated, or did not change from the time of the

SWU. The clinical condition deteriorated in 10 of the

85 episodes, but PCT and CRP values at the time of

the sepsis workup (0.26 ng/ml (interquartile range

3.62) and 0.86 mg/l (1.1)) were similar to the

episodes where infants improved (0.63 ng/ml (1.86)

and 1.34 mg/l (1.12); Mann-Whitney test, p�/0.2 and

0.7, respectively). PCT levels did not differ at the

other time intervals.

Associated factors

Six infants (all from the septic groups) in whom

SWU was associated with respiratory distress

(FiO2�/0.5) had increased PCT values, independent

of other factors, including sepsis status (multiple

linear regression analysis, p B/0.001). The presence

of a patent ductus arteriosus (n�/ 24), necrotizing

enterocolitis (n�/10: seven in group 3, two in group

2, and one in group 1), pneumonia (n�/4), and

cellulitis (n�/4) did not contribute independently to

PCT level. The number of infants with intraventri-

cular hemorrhage, hypertension, convulsions, and

renal failure was too small to enable statistical

correlation.

Discussion

In this prospective study, we monitored a cohort of

preterm infants who underwent SWU during their

hospitalization, and compared PCT and CRP levels,

in those who were subsequently defined as septic, with

those of infants who proved not to have sepsis. A

significant increase in PCT was identified in the septic

groups during the first 48 h after the SWU. The area

under the ROC curve (0.74), however, indicates that

Table II. Sensitivity, specificity and predictive values of procalcitonin (PCT, ng/ml) and C-reactive protein (CRP, mg/l) in detecting sepsis.

n Sensitivity, % (95% CI) Specificity, % (95% CI) PPV, % (95% CI) NPV, % (95% CI)

At the time of sepsis workup

PCT�/0.5 35 74 (52�87) 54 (38�70) 53 (36�70) 78 (58�89)

PCT�/1.0 20 48 (30�68) 88 (73�95) 73 (58�84) 73 (48�89)

PCT�/2.3 15 48 (30�68) 97 (85�99) 91 (65�99) 74 (61�85)

CRP�/10 56 74 (58�87) 39 (25�52) 46 (33�59) 68 (48�83)

CRP�/20 32 47 (31�64) 89 (77�95) 75 (53�89) 70 (57�80)

CRP�/30 15 41 (26�58) 96 (85�99) 87 (62�96) 69 (57�79)

CRP�/50 11 31 (18�49) 98 (88�99) 91 (62�98) 67 (55�77)

CRP�/10 or PCT�/0.5 60 93 (76�98) 21 (11�36) 44 (31�57) 80 (49�94)

At 25�72 h after sepsis workup

PCT�/0.5 15 67 (42�85) 62 (41�79) 56 (34�75) 72 (49�88)

PCT�/1.0 14 60 (36�80) 81 (60�92) 69 (42�87) 74 (54�87)

PCT�/2.3 9 47 (25�69) 95 (78�99) 88 (53�98) 71 (54�85)

n represents number of samples, without repeated measurements; PPV/NPV: positive/negative predictive value.

0.18.06.04.02.00.0

yticificepS - 1

0.0

2.0

4.0

6.0

8.0

0.1

TCPPRC

Sens

itiv

ity

Figure 2. Receiver operating characteristic (ROC) curve of procal-

citonin (PCT) and C-reactive protein (CRP) in detecting nosoco-

mial sepsis in preterm infants, at the time of the sepsis workup. Area

under the curve for PCTand CRP is 0.74 (95% CI 0.59�0.88) and

0.73 (0.59�0.84), respectively.

1574 D. Turner et al.

this test is only fair in discriminating septic from non-

septic infants, but comparable to CRP (0.73). Using a

high cutoff level, both procalcitonin (2.3 ng/ml) and

CRP (30 mg/l) had high specificity and positive

predictive value (97%, 91% and 96%, 87%, respec-

tively), but low sensitivity (48% and 41%, respec-

tively). The high specificity was achieved despite the

contribution of severe hypoxia to PCT values, as also

suggested by others [3].

The clinical decision as to when to discontinue

antibiotic therapy in a neonate with negative

cultures and who has become asymptomatic is often

a difficult one. A high PCT (�/2.3 ng/ml) or CRP

(�/30 mg/l) value suggests that antibiotics should be

continued. In our study, 14 of the 22 episodes (64%)

with PCT�/2.3 ng/ml were culture positive, while six

of the remaining were classified in the non-septic

groups. On the other hand, combined levels of CRPB/

10 mg/l and PCTB/0.5 ng/ml yielded a negative

predictive value of 80%, indicating that sepsis is less

likely.

The main advantage of LR is that it surpasses

the simple classification of a test result as either

normal or abnormal [14]. For instance, the pretest

probability of sepsis in our study (calculated as

40% by the proportion of proven sepsis out of the

total SWU episodes) will increase to 95% if the

PCT value is �/3 ng/ml (LR 22.8, using Fagan’s

nomogram [14]) and decrease to 25% if the value is

0�1 ng/ml (LR 0.5).

Two previous studies evaluated the role of PCT in

the detection of nosocomial sepsis in preterm infants,

the first of which [7] showed both high specificity

(100%) and sensitivity (100%). The second study [8],

in contrast, used a PCT cutoff of 0.5 ng/ml, achieving

high sensitivity (97%) but lower specificity (80%).

These studies used healthy infants as the comparison

group, thereby obtaining better accuracy of the test.

By contrast, in our study, the comparison group was

composed of ill-appearing premature infants who

required a sepsis workup, but who were eventually

found not to be septic. In the clinical setting, this

difference is crucial, as diagnostic laboratory tests are

necessary to exclude sepsis in ill infants and not in

healthy ones.

The strengths of this study lie, therefore, in its

prospective design, and in the use of a comparison

group of ill infants who eventually proved not to be

septic. A limitation of our study is the small number of

blood samples drawn from each infant. This restric-

tion was governed by the ethical decision to obtain

blood samples only as part of routine venipuncture.

Nevertheless, each PCT sample contributed indepen-

dently to the analysis at separate time intervals,

and the high enrollment rate also compensates for

this weakness. An additional limitation to consider

is the lack of a gold standard to diagnose sepsis,

which may have led to measurement bias and type 2

errors. We used stringent criteria for sepsis definition

in order to minimize this potential bias, but one

cannot rule out an underestimation of the perfor-

mance of PCT.

In conclusion, PCT level is elevated during noso-

comial sepsis in preterm infants, and its overall

performance is comparable to CRP. PCT levels of

�/2.3 ng/ml and CRP �/30 mg/l are sufficiently

specific to continue antibiotic therapy despite sterile

cultures. The introduction of likelihood ratios can

assist in calculating the probability of sepsis in an

individual infant.

Acknowledgements

BRAHMS Diagnostica (Berlin, Germany) provided

the testing kits for PCT determination but was not

involved in any aspect of the study and manuscript

preparation. The authors thank the nurses of the

NICU who meticulously collected the blood samples,

and Prof. P. M. Sherman and Prof. A. Eidelman for

their aid in preparing the manuscript.

References

[1] Kaufman D, Fairchild KD. Clinical microbiology of bacterial

and fungal sepsis in very-low-birth-weight infants. Clin

Microbiol Rev 2004;/17:/638�80.

[2] Chiesa C, Pacifico L, Mancuso G, Panero A. Procalcitonin in

pediatrics; overview and challenge. Infection 1998;/26:/236�41.

[3] van Rossum AM, Wulkan RW, Oudesluys-Murphy AM.

Procalcitonin as an early marker of infection in neonates and

children. Lancet Infect Dis 2004;/4:/620�30.

[4] Assicot M, Gendral D, Carsin H, Raymond J, Guilbaud J,

Bohuon C. High serum PCT concentration in patients with

sepsis and infection. Lancet 1993;/341:/515�8.

[5] Meisner M, Schmidt J, Huttner H, Tschaikowsky K. The

natural elimination rate of procalcitonin in patients with

normal and impaired renal function. Intensive Care Med

2000;/26 Suppl 2:/P212�6.

[6] Distefano G, Curreri R, Betta P, Romeo MG, Amato M.

Procalcitonin serum levels in perinatal bacterial and fungal

infection of preterm infants. Acta Paediatr 2004;/93:/216�9.

[7] Chiesa C, Pacifico L, Rossi N. Procalcitonin as a marker of

nosocomial infections in the neonatal intensive care unit.

Intensive Care Med 2000;/26 Suppl 2:/P175.

Table III. Interval likelihood ratios (LR) of procalcitonin (PCT, ng/

ml) and C-reactive protein (CRP, mg/l) in detecting sepsis, at the

time of the sepsis workup.

CRP LR 95% CI PCT LR 95% CI

0�10 0.6 0.3�1.2 0�1.0 0.5 0.3�0.8

11�30 0.6 0.4�1.1 1.1�2.0 1.4 0.1�14

31�80 7.1 0.9�57.3 2.1�3.0 8.6 0.9�78

�/80 11.3 1.5�85.6 �/3 22.8 3.1�171

Procalcitonin in preterm infants 1575

[8] Vazzalwar R, Pina-Rodrigues E, Puppala BL, Angst DB,

Schweig L. Procalcitonin as a screening test for late-onset

sepsis in preterm very low birth weight infants. J Perinatol

2005;/25:/397�402.

[9] Chiesa C, Panero A, Rossi N, Stegagno M, De Giusti M,

Osborn JF, et al. Reliability of procalcitonin concentrations for

the diagnosis of sepsis in critically ill neonates. Clin Infect Dis

1998;/26:/664�72.

[10] Monneret G, Labaune JM, Isaac C, Bienvenu F, Putet G,

Bienvenu J. Procalcitonin and C-reactive protein levels in

neonatal infections. Acta Paediatr 1997;/86:/209�12.

[11] Lloyd BW, Oto A. Normal values for mature and immature

neutrophils in very preterm babies. Arch Dis Child 1982;/57:/

233�5.

[12] Oski F, Naiman J. Hematologic problems in the newborn.

Philadelphia: WB Saunders; 1982.

[13] Roberts WL, Moulton L, Law TC, Farrow G, Cooper-

Anderson M, Savory J, et al. Evaluation of nine automated

high-sensitivity C-reactive protein methods: implications for

clinical and epidemiological applications. Part 2. Clin Chem

2001;/47:/418�25.

[14] Jaeschke R, Guyatt GH, Sackett DL. Users’ guides to the

medical literature. III. How to use an article about a diagnostic

test. B. What are the results and will they help me in caring for

my patients? The Evidence-Based Medicine Working Group.

JAMA 1994;/271:/703�7.

1576 D. Turner et al.