Received: 14 August 2018 

|   Revised: 20 June 2019 

|  Accepted: 8 July 2019

DOI: 10.1111/crj.13062

ORIGINAL ARTICLE

Plasma cytokine profile on admission related to aetiology in

community‐acquired pneumonia

Eduard H. Burgmeijer1   | Ruud Duijkers1  | René Lutter2  | Marc J. M. Bonten3  |

Valentijn A. Schweitzer   3
| Wim G. Boersma 1

1
Department of Pulmonology, North West
Hospital Alkmaar, Alkmaar,
Abstract
the Netherlands Background: Potentially unnecessary antibiotic use for community‐acquired pneu-
2
Departments of Respiratory Medicine and monia (CAP) contributes to selection of antibiotic‐resistant pathogens. Cytokine ex-
Experimental Immunology, Amsterdam
pression at the time that treatment is started may assist in identifying patients not
UMC, University of Amsterdam,
Amsterdam, the Netherlands requiring antibiotics. We determined plasma cytokine patterns in patients retrospec-
3
Department of Molecular Epidemiology of tively categorized as strict viral, pneumococcal or combined viral‐bacterial CAP.
Infectious diseases, Department of Medical Objective: To investigate whether cytokine‐based prediction models can be used to
Microbiology and Julius Centre for Health
Sciences and Primary Care, University
differentiate strict viral CAP from other aetiologies at admission.
Medical Centre Utrecht, Utrecht, Methods: From 344 hospitalized CAP patients, 104 patients were categorized as
the Netherlands viral CAP (n = 17), pneumococcal CAP (n = 48) and combined bacterial‐viral CAP
Correspondence (n = 39). IL‐6, IL‐10, IL‐27, IFN‐γ and C‐reactive protein (CRP) were determined
Wim G. Boersma, Department of on admission in plasma. Prediction of strict viral aetiology was explored with two
Pulmonology, North West Hospital
multivariate regression models and ROC curves.
Alkmaar, Wilhelminalaan 12, NL1815‐JD,
Alkmaar, The Netherlands. Results: Viral pneumonia was predicted by logistic regression using multiple cy-
Email: [email protected] tokine levels (IL‐6, IL‐27 and CRP) with an AUC of 0.911 (95% CI: 0.852‐0.971,
P < .001). For the same patients the AUC of CRP was 0.813 (95% CI: 0.728‐0.898,
P < .001).
Conclusions: This study demonstrated differences in cytokine expression in selected
CAP patients between viral and bacterial aetiology. Prospective validation studies
are warranted.

KEYWORDS
cytokines, immunology, pneumonia, viral infection

1  |   IN T RO D U C T ION world and is often empirically treated with antibiotics.1-3

Despite the use of several time‐consuming microbial and
Of all infectious diseases, community‐acquired pneumonia molecular techniques, a conclusive microbiological diagno-
(CAP) is the number one cause of death in the developed sis is only established in up to 50% of patients presenting

Abbreviations: 95%‐CI, 95% confidence interval; AUC, area under the curve; BAL, broncho alveolar Lavage; CAP, community‐acquired pneumonia;
COPD, chronic obstructive pulmonary disease; CRP, C‐reactive protein; EDTA, ethylenediaminetetraacetic acid; ELISA, enzyme‐linked immunosorbent
assay; ICU, intensive care unit; IFN, interferon; IL, interleukin; IP, interferon‐inducible protein; LLOD, lower limit of detection; LOS, length of stay; NK,
natural killer (cells); NPV, negative predictive value; PCR, polymerase chain reaction; PCT, procalcitonin; PPV, positive predictive value; RCT, randomized
controlled trial; ROC, receiver operating characteristic; RSV, respiratory syncytial virus; TNF, tumour necrosis factor.

Clin Respir J. 2019;13:605–613. wileyonlinelibrary.com/journal/crj © 2019 John Wiley & Sons Ltd     605 |
 |
606       BURGMEIJER et al.

with CAP.4-7 Respiratory viruses can be identified by poly- In the present study, we investigated whether plasma
merase chain reaction (PCR) in 20%‐40% of CAP patients.6 cytokine levels of the Th17, Th1 and Th2 pathways can be used
If respiratory pathogens are detected by PCR, these viral to distinguish between three aetiological groups: pneumococ-
agents can be causative or non‐causative for infection. When cal, viral and mixed viral/bacterial infection. First, we inves-
causative, they can be in fact coinfections with undetected tigated absolute cytokine plasma‐level differences between
bacterial pathogens or strict viral infection. ‘Strict viral’ CAP groups. Hereafter, we investigated whether cytokine‐based
is defined as when a respiratory virus is the only causative prediction models can be used to differentiate viral CAP from
pathogen for CAP in a patient. For strict viral CAP, antibiot- other aetiologies at admission, and whether this adds value to
ics are probably ineffective and in theory should be withheld. the routine determination of CRP.
Symptom‐based prediction of aetiology has proven inade-
quate to discriminate between viral and bacterial aetiology.8-11
An alternative strategy to predict bacterial aetiology is the
2  |  M ETHODS
use of biomarkers like C‐reactive protein (CRP) or procal-
2.1  | Patients
citonin (PCT).12,13 High (>0.5 mcg/L) PCT values are used
to initiate the use of antibiotics and low values (<0.1 mcg/L) In the present study, patients were admitted to the Northwest
strongly discouraged the use of antibiotics. However, this Hospital group Alkmaar with CAP from October 2013 to
strategy fails to distinguish viral from atypical pathogens and September 2016. Patients were part of the REDUCE trial,
thus cannot be used to identify those patients in which antibi- which was a randomized controlled multicentre trial in pa-
otics can be withheld.14 tients admitted to a regular hospital ward with radiologically
In the early phase of pneumonia, bacteria and viruses proven CAP (NCT01964495). All data were prospectively
trigger distinct innate immune response pathways. As a con- collected in a standardized manner. Blood samples, blood
sequence, several differentiating inflammatory mediators cultures for aerobic and anaerobic pathogens, oropharyngeal
are likely to be markedly elevated and could serve as po- swabs for respiratory viruses and atypical pathogens, urine
tential biomarkers. In bacterial CAP, rapid interleukin‐17A antigen tests for Legionella pneumophila serotype 1 and
(IL‐17A) production by gamma‐delta T cells attracts, ex- S. pneumoniae and if possible, sputum cultures were obtained
pands and activates neutrophils at the site of infection.15 for all patients. The regional ethical committee approved the
Release of young neutrophils from the bone marrow is an REDUCE trial, and the re‐use of samples for this retrospec-
innate response aimed at mainly extracellular pathogens, tive studies.
such as pneumococci.16 IL‐6 enhances general pro‐inflam-
matory activity as well as T‐helper‐17 (Th‐17) development
2.2  |  Inclusion and exclusion criteria
from naïve T cells.17 In bacterial pneumonia higher levels of
IL‐6, TNF‐α and interleukin‐1 (IL‐1) were found in broncho- Patients were eligible for inclusion in the study if they met
alveolar lavage (BAL) compared to healthy controls. Also the following inclusion criteria: age ≥18, need for hospitali-
in serum, IL‐6 was elevated in bacterial CAP compared to zation and a life expectancy >30 days, informed consent was
healthy individuals.18 A certain ‘spill’ of cytokines or a sys- obtained from either the patient or their legal representative.
temic response might be responsible for this finding. All patients had to have new consolidation(s) on the chest
In viral CAP, type 1 interferons (IFN) are produced by radiograph and a clinical presentation of an acute illness with
infected cells, often airway epithelial cells. Natural Killer one or more of the following symptoms: temperature ≥38.0°C
(NK) cell and also CD8 T cells will produce type 2 interferon (100.4°F), dyspnoea, cough (with or without expectoration
(IFN‐γ) in response to viral replication. Interferons limit viral of sputum), chest pain, malaise or fatigue, myalgia, gastroin-
replication and enhance the T‐helper‐1 pathway response. testinal symptoms, rales, rhonchi or wheezing, egophony or
However, T‐helper‐2 pathway cytokines, such as IL‐5, results bronchial breath sounds and haemoptysis. Exclusion criteria
in the recruitment and activation of eosinophils, which can were: severe immunosuppression (eg, HIV infection, chemo-
display anti‐viral activities as well.19,20,21 therapy), active neoplastic disease, obstruction pneumonia
In mixed bacterial/viral infections, primary viral patho- (eg, in lung cancer), aspiration pneumonia, pneumonia that
gens can enhance bacterial infection. Adherence of bacteria to developed within 8  days after hospital discharge, inability
epithelial cells is enhanced in virus‐infected cells, predispos- and/or unlikeliness to comprehend and/or follow the proto-
ing to superinfection,22 but several alternative mechanisms col, pregnancy and/or lactation.
have been proposed. CAP patients with primary influenza
infection have elevated levels of IL‐27, which in turn inhibit
2.3  |  Categories of CAP
IL‐10 production and therewith the Th‐17 pathway. This does
not alter viral clearance, but limits lung neutrophil influx and We defined three aetiological groups based on pathogens
potentially diminishes bacterial clearance.23-25 demonstrated by routine microbiological procedures. Patients
BURGMEIJER et al.   
|
   607

in which no pathogens were detected or patients that did not (unilobar, multilobar, bilateral). CURB‐65 scores vali-
match the criteria for one of the three groups were excluded dated to predict short‐term mortality for CAP were calcu-
from the analysis (Figure 1). Groups were defined as: strict lated. Primary outcome variables consisted of the following
viral CAP (n = 17); oropharyngeal swab PCR‐positive for a cytokines in picogram per millilitre (pg/mL): IL‐1β, IL‐4,
virus, PCT levels on admission ≤0.25 μg/L and no bacteria IL‐5, IL‐6, IL‐10, IL‐12p70, IL‐17A, IL‐21, IL‐22, IL‐23,
detected. Adenoviruses and rhinoviruses were excluded from IL‐27, TNF‐α, IFN‐γ and IP‐10. Secondary outcome variable
this group as they are generally not considered to be virulent was the CRP serum level upon admission in milligram per
enough to cause CAP in this sample of adult immunocom- litre (mg/L).
petent patients with community‐based aetiology. We believe
that patients having a sole adenovirus or rhinovirus sampled,
2.5  |  Cytokines assay
cannot be stated as viral CAP with enough certainty to in-
clude. Pneumococcal CAP (n = 48); pneumococci detected Luminex assay (eBioscience, ProcartaPlex) was used to
by any microbial technique (blood, sputum, urine antigen measure cytokines plasma levels, which were read on a
test) and no other pathogens identified. Mixed CAP (n = 39): Bioplex 200 (BioRad). Samples were collected and processed
defined by the presence of at least one bacterial species plus in a standardized manner on admission and stored at −80°C
the presence of at least one viral species (other than adenovi- at the laboratory of the Northwest Hospital till analyses at the
rus and rhinovirus) by PCR, and no fungi or yeasts detected. Academical Medical Centre of Amsterdam. Samples had not
been thawed before.
2.4  |  Baseline and outcome measurements
2.6  | Statistics
Baseline characteristics comprised age, gender, co‐morbidity
according to the Charlson index, smoking status (current/ Descriptive statistics are used to compare baseline char-
non‐current), chest radiograph localization of consolidation acteristics between strict viral, pneumococcal and mixed

F I G U R E 1   Numbers of patients
per group. Final three groups are
shown underlined. Patients in which no
pathogen was detected, patients with non‐
pneumococcal bacterial aetiology, single
adeno/rhinovirus infected patients were
excluded from analysis. Viral CAP patients
with PCT ≥ 0.25 μg/L (n = 10) were
excluded because of possible coinfection
with bacterial pathogen(s)
 |
608       BURGMEIJER et al.

CAP. Continuous variables are presented as mean or me- pulmonary disease (COPD). The various pathogens per
dian with standard deviation or IQR depending on the aetiological group are listed in Table 2.
normality of the distribution and categorical variables as There were significant differences in median plasma
proportions. Differences in cytokine levels between the levels of IL‐6, IL‐10, IL‐17A and IFN‐γ between the three
three groups were tested with the Kruskal‐Wallis test, fol- aetiologies (Table 3). IFN‐γ was elevated in viral CAP pa-
lowed by Mann‐Whitney tests between groups where ap- tients. IL‐6 was elevated in pneumococcal CAP. These two
propriate. For the prediction model, we compared the strict cytokines were significantly different in all three groups even
viral CAP with the remaining CAP patients (pneumococcal though there was still some overlap as depicted in Appendix
CAP and mixed CAP). First, we constructed a model con- Figure A‐1. In order to explore if prediction of aetiology
taining only cytokines, then containing only CRP and then was possible, we created two prediction models using mul-
compared this to a model with cytokines and CRP together. tivariate logistic regression: the first model consisted only of
Last, we created two models on a population without selec- the measured cytokines, the second one also included CRP
tion by PCT. (Table 4). In the models (IFN‐γ)2 was used because IFN‐γ
We performed multivariate logistic regression with back- was collinear with the outcome, all other cytokines were not.
ward selection using the Wald statistic with a P value of .1, To prove the clinical use of cytokine measurements over
starting with all cytokines and adding CRP as a fixed vari- CRP alone, we compared to an AUC of CRP alone, which
able. The predicted probabilities for having strict viral CAP has an AUC of 0.813 (95% CI: 0.728‐0.898, P < .001). ROC
were calculated for each individual subject, and analysed curves were created for both models (Figure 2). IL‐5 and
with ROC curves. Linearity of cytokines to the log odds of IL‐27 contributed to the models, although these cytokines
the outcome were assessed using Box‐Tidwell tests. IBM were non‐significant as single predictors of aetiology.
SPSS Statistics for Windows, version 24.0 was used for all In a cytokine‐alone model, IL‐5, IL‐6 and IFN‐γ appeared
analyses. A P value below .05 was considered statistically to distinguish viral from bacterial CAP (ie, pneumococcal
significant for all data but the selection of variables in the and mixed CAP together) best. In this first model, 35% of the
prediction models. variance in subjects either having strict viral or other aetiol-
ogy CAP, can be explained by the cytokines IL‐5, IL‐6 and
IFN‐γ (R2 = .353). Figure 2 shows the diagnostic accuracy to
3  |   R ES U LTS differentiate between strict viral and bacterial‐induced CAP.
The cytokine‐based model calculates the chance of viral CAP
The demographics of the three groups did not reveal sta- for each CAP patient. We created an AUC curve with an area
tistically significant differences in gender, smoking status, under the curve of 0.863 (95% CI: 0.768‐0.958, P < .001).
comorbidities and disease severity (Table 1). All groups con- Using a cut‐off value of 0.65, there is 99% specificity and
tained comparable rates of patients with chronic obstructive 18% sensitivity.

T A B L E 1   Baseline variables
Strict viral Pneumo coccal Mixed Infection
(n = 17) (n = 48) (n = 39)
Age in years 71 (17) 68 (15) 67 (19)
Male gender 7/17 (41%) 28/48 (58%) 17/39 (44%)
Median Charlson index 1 1 0
COPD 2/17 (12%) 17/48 (35%) 10/39 (26%)
Current smoker 2/17 (12%) 14/45 (31%) 9/39 (23%)
Consolidation
unilobar 10/15 (67%) 39/47 (83%) 26/34 (76%)
multilobar 5/15 (33%) 8/47 (17%) 8/34 (24%)
bilateral 1/15 (7%) 7/47 (15%) 5/34 (15%)
CURB‐65 score
1 3/17 (18%) 12/48 (25%) 14/39 (36%)
2 9/17 (53%) 11/48 (23%) 9/39 (23%)
3 3/17 (18%) 13/48 (27%) 11/39 (28%)
Notes: Age is displayed in median (IQR). All other variables are displayed in n/group total (%). All patients
had CURB‐65 1, 2 or 3.
BURGMEIJER et al.   
|
   609

T A B L E 2   Detection of pathogens per aetiological group

Pneumococcal
CAP (n = 48) Sputum culturea Blood culture Urinary antigen test
S. pneumoniae 13/25 (52%) 18/48 (38%) 30/48 (63%)
Strict viral patho- Oropharyngeal swab
gens (n = 17)
Influenza A/B virus 7/17 (41%)
RS virus 4/17 (24%)
Parainfluenza virus 3/17 (18%)
Corona virus 2/17 (12%)
Human 1/17 (6%)
metapneumovirus
Bocavirus 0/17 (0%)
Mixed pathogens
(n = 39)b
Influenza RSV PI CV hMPV
S. pneumoniae 9/39 (23%) 3/39 (8%) 2/39 (5%) – 3/39 (8%)
H. parainfluenzae 6/39 (15%) – 1/39 (3%) – 3/39 (8%)
H. influenzae 4/39 (10%) 1/39 (3%) – 2/39 (5%) –
M. catarrhalis 3/39 (8%) – – 1/39 (3%) 1/39 (3%)
S. aureus 1/39 (3%) 1/39 (3%) 1/39 (3%) – 1/39 (3%)
E. coli 1/39 (3%) 1/39 (3%) 1/39 (3%) – –
K. pneumoniae – – – – 1/39 (3%)
K. oxytoca – – – – 1/39 (3%)
S. marcescens 1/39 (3%) – – – –
Notes: Relative number (percentage) of patients in which the microorganism was detected per aetiological group. Influenza, influenza virus A or B; RSV, respiratory
syncytial virus; PI, parainfluenza virus; CV, corona virus; hMPV, human metapneumovirus.
a
In the pneumococcal group, for (48‐25=) 23 patients no sputum sample was obtained.
b
In the mixed aetiology group, for (39‐32=) 7 patients no sputum sample was obtained.

T A B L E 3   Levels of plasma cytokines

Aetiology
in the three CAP groups
Strict viral Pneumococcal
Cytokine (pg/mL) n = 17 n = 48 Mixed n = 39 P value LLOD
IL‐1β 1.6 (2.12) 1.6 (1.95) 1.6 (1.56) .960 0.167
IL‐4 0.00 (2.38) 0.00 (0.47) 0.00 (0.47) .591 0.639
IL‐5 2.21 (3.95) 1.22 (2.04) 1.22 (1.51) .401 0.352
IL‐6 12.9 (55.5) 378 (1444) 66.6 (288) <.001 2.03
IL‐10 5.3 (10.8) 12.1 (30.4) 6.78 (28.5) .048 0.090
IL‐12p70 0.00 (0.00) 0.00 (0.10) 0.00 (0.04) .135 0.397
IL‐17A 0.00 (0.00) 0.04 (0.25) 0.00 (0.04) .002 0.101
IL‐23 0.00 (0.13) 0.00 (0.26) 0.00 (0.26) .207 0.733
IL‐27 0.00 (10.7) 0.00 (17.6) 0.00 (28.3) .248 5.35
IFN‐γ 4.88 (12.3) 0.97 (1.70) 1.45 (1.45) .002 0.305
TNF‐α 0.00 (0.56) 0.00 (0.29) 0.03 (0.50) .531 0.249
IP‐10 91.8 (210) 78.5 (142) 116 (233) .484 0.560
Notes: Median (IQR) cytokine concentrations per group in picogram/ml. For IL‐17A, median plasma levels are
just below Lower limit of detection (LLOD) of the Luminex assay, but reliably extrapolated from the calibra-
tion curve in 19 subjects. IL‐21, IL‐22 and TNF‐α are not displayed as they were undetectable in all samples.
 |
610       BURGMEIJER et al.

T A B L E 4   Two created cytokine prediction models, with and shows that the ROC curve of the predicted probabilities from
without CRP all subjects with viral CAP, had an AUC of 0.911 (95% CI:
95% confi-
0.852‐0.971, P < .001). Using a cut‐off value of 0.65, a spec-
B OR dence interval ificity of 99% and sensitivity of 35% was found. In compar-
ison, prediction by CRP alone would have an AUC of 0.813
Cytokine prediction model
(95% CI 0.728‐0.898, P < .001).
IL‐5 (pg/mL) 0.076 1.079 0.988‐1.178
When doing a sensitivity analysis on all viral CAP pa-
IL‐6 (pg/mL) −0.008 0.992 0.986‐0.999 tients without using PCT as a selection criterion (n  =  27),
IFN‐γ (pg/mL) 0.053 1.055 0.992‐1.121 we created two extra ROC curves with lower AUC values.
Constant −1.156 0.315 A cytokine‐only prediction model had an AUC of 0.726 (CI:
Cytokine and CRP prediction model 0.624‐0.829, P  <  .001). The prediction model using both
IL‐6 (pg/mL) −0.008 0.992 0.986‐0.999 cytokines and CRP has an AUC of 0.775 (CI: 0.681‐0.870,
IL‐27 (pg/mL) −0.050 0.951 0.907‐0.997 P < .001).
CRP (mg/L) −0.018 0.982 0.971‐0.993
Constant 1.892 6.629
Notes: Top panel of the table: cytokine‐based prediction model. Bottom panel:
4  |  DISCUSSION
cytokine plus CRP‐based prediction model. Odds ratio’s in these models are the
odds of having strict viral CAP compared to the odds of having pneumococcal
In the present study, systemic cytokine levels differed be-
or mixed CAP, for every unit change in plasma level of that cytokine. B, regres- tween patients classified upon documented microbiological
sion coefficient per unit change in plasma level; OR, Odds ratio per unit change cause of CAP. Differences were observed for IL‐6, IL‐10,
in plasma level; 95% confidence interval for given odds ratio’s. IL‐17A and IFN‐γ, yet with considerable overlap for all cy-
tokines between all groups. Cytokine measurements may add
When adding CRP in the selection method, the cytokines predictive value to the routine clinical measurements of CRP.
IL‐6, IL‐27 and CRP appeared to distinguish viral from other However, current findings cannot be extrapolated to an un-
aetiologies of CAP. In this model, 59% of the variance in sub- selected cohort of patients, thereby being merely a proof of
jects either having either strict viral or the other two aetiol- concept. These findings need prospective validation in unse-
ogies of CAP (mixed and pneumococcal) can be explained lected patients to determine predictive values and the effects
by the cytokines IL‐6, IL‐27 and CRP (R2 = .587). Figure 2 on patient management and patient outcome.

IL-6 IL-27 CRP prediction model

CRP

AUC 0.911 (95% CI: 0.852 – 0.971, p<0.001)

F I G U R E 2   ROC curves of both multiple cytokine prediction models for distinction between strict viral and the combined pneumococcal and
mixed aetiology group, using two multiple cytokine prediction models
The left picture displays the first (cytokine‐alone) model. The right figure displays the second (cytokines + CRP) model in blue (AUC 0.911
(95% CI: 0.852‐0.971, P < .001) and prediction by CRP alone in green as a reference (AUC of 0.813 (95% CI 0.728‐0.898, P < .001).
BURGMEIJER et al.   
|
   611

Previous studies have focused on individual biomarkers to the unknown‐aetiology group. Only patients with a defi-
to differentiate aetiology in CAP. The best‐studied biomarker nite microbial diagnosis were selected for our model, and in
is the plasma IL‐6 level. IL‐6 proved to differentiate between general a definite microbiological diagnosis is established in
typical and viral CAP, between pneumococcal CAP and only 40%‐50% of patients presenting with CAP.4-7
Mycoplasma pneumoniae and between pneumococcal and Furthermore, group definitions may influence results. We
non‐pneumococcal CAP respectively.26-28 Interestingly, CAP choose to exclude adenoviruses and rhinoviruses because of
severity scores and mortality have been found to correlate the uncertainty to assign them to be the causative pathogen in
well with cytokine IL‐6 levels on the first day of hospital- patients in which this is the only pathogen found. We choose
ization.29,30 This poses the question whether cytokine levels the pneumococcal CAP group for its homogeneity, but did
are specific for aetiology or specific for disease severity, or not select other bacterial CAP groups, because these groups
both. Menendez et al.7 demonstrated that peripheral IL‐6 is would be too small for meaningful statistical analysis. We ex-
elevated in CAP presenting with acute sepsis or shock de- pected the mixed CAP group to be heterogeneous, both in
spite the aetiology, but is also elevated in CAP caused by aetiology as in cytokine expression. Subgroup analysis was
Gram‐positive cocci without septic shock. Endeman et al.26 not reliable with a maximum of nine subjects having compa-
reported that IL‐6 levels in blood predict pneumococcal CAP rable co‐infection (influenza‐pneumococci). Noteworthy, is
when independent of age and PSI. In our analysis, CURB‐65 the absence of atypical pathogens in our mixed CAP group,
score alone or added to our multivariable models did not im- despite the use of routine PCR on oropharyngeal swab, which
prove prediction of aetiology (data not shown). Unlike other increases detection rate.33
studies, patients directly admitted to the ICU were excluded We did not take pretreatment with antibiotics or predni-
in the REDUCE study, consequently data of these more se- sone prior to admission into account. Previous studies stated
verely ill patients are not available. that antibiotic pretreated patients had lower IL‐10 and IL‐6
The major limitation of the present study is that the anal- levels, compared to treatment‐naïve patients.7,28 Endeman
yses were retrospective (although the data and samples were et al.26 reported that patients treated with corticosteroids on
prospectively collected), with extremely well‐selected and de- admission or prior to admission had significantly lower IL‐6
fined patient groups and with exclusion of patients in which no levels on day 3 compared to patients not treated with corti-
pathogen was detected. We used a PCT cut‐off of <0.25 μg/L costeroids at all. Furthermore, we are unsure about the effect
to select a group in which it was highly unlikely that a bacte- of COPD on our results, since chronic lung inflammation
rial co‐infection was present. We believe this is necessary as a in COPD patients may alter the immune response towards
criterion to exclude possibly bacterial and viral co‐infections, a Th‐1 direction.34 The pitfall of CAP in COPD patients is
since the microbiological techniques used are not sensitive that causality between pathogen detection and disease is chal-
and specific enough to reduce this possible bias. Prospective lenging. For example, 25% of ambulant COPD patients carry
validation in an unselected cohort is needed to validate our clinically significant amounts of pathogens in the lower air-
results and compare them with other diagnostic strategies to ways while not suffering from an exacerbation, compared to
withhold or withdraw antibiotics, such as solely PCT or CRP. 52% potential pathogens during exacerbation.35
In a sensitivity analysis, we saw CRP alone predicted less than Timing of measurement may have influenced the results
CRP and cytokines combined. This suggests cytokines add as well. Cytokine levels generally decline in the course of
value to the prediction of viral CAP. In a second sensitivity disease, influence other pathways or alter after treatment is
analysis, we excluded the use of PCT as a selection criterion. initiated.26,36 These factors should be taken into account in
The then‐created ROC curves for the correlation predicted future research aimed at further validating the prediction
worse, with lower AUC values. Thus, we believe PCT may model. Technical aspects might have influenced our results as
be needed as a selection criterion in order to exclude unde- well. TNF‐alpha, for example, was low or even undetectable.
tected bacterial pathogens from the viral group, which creates Collection of samples was highly standardized (rapid process-
a model predicting better than without the PCT criterion. ing after blood withdrawal) and analysis were performed by
Choosing a PCT criterion lowers, but not completely rules experienced personnel in a highly standardized setting and
out, the chance of including individuals with undetected bac- appropriate controls were used. Some technical variation in
terial pathogens to our strict viral CAP group.31 Low PCT detection, for example, by the use of different Luminex pan-
was found to differentiate typical from atypical CAP, but not els, might have also influenced our TNF‐alpha results.
atypical from viral CAP.14,32 So, atypical pathogens that were Rapid cytokine tests such as Simple Plex are found to
not detected by PCR, could potentially have been selected in be as accurate as Luminex, but require less time and human
our presumably strict viral group. effort.37 Future research is needed to investigate the benefit
Exclusion of patients with an indefinite microbial cause and pitfalls of a rapid, cytokine‐based prediction model and
may create considerable bias. Menendez et al.7 reported if this predicted diagnosis is reliable enough to base treat-
higher IL‐6 levels in the known‐aetiology group compared ment strategy on. When validating this studies’ prediction
 |
612       BURGMEIJER et al.

model prospectively, up to half of all patients will likely not management of community‐acquired pneumonia in adults. Neth J
have a definite microbial diagnosis. Before implementing Med. 2012;70(2):90‐101.
an eventual strategy to withhold antibiotics based on this 2. Lee JS, Giesler DL, Gellad WF, Fine MJ. Antibiotic therapy for
adults hospitalized with community‐acquired pneumonia: a sys-
model, it should be validated in prospective trials and a cost‐
tematic review. JAMA. 2016;315(6):593‐602.
effectiveness analysis should be performed. Several outcome 3. Mandell LA, Wunderink RG, Anzueto A, et al. Infectious Diseases
variables such as length of stay, treatment failure and mortality Society of America/American Thoracic Society consensus guide-
should be documented. Finally, costs of cytokine assays and lines on the management of community‐acquired pneumonia in
analysts performing it should outweigh the consequences of adults. Clin Infect Dis. 2007;44(suppl 2):S27‐S72.
antibiotic resistance, for this approach to be cost‐efficient. 4. Musher DM, Roig IL, Cazares G, Stager CE, Logan N, Safar H.
In the present retrospective cohort analysis, we found Can an etiologic agent be identified in adults who are hospitalized
for community‐acquired pneumonia: results of a one‐year study.
pathogen specific differences in cytokine levels of CAP pa-
J Infect. 2013;67(1):11‐18.
tients on admission. Combination of predictive value of cyto-
5. van der Eerden MM, Vlaspolder F, de Graaff CS, et al. Comparison
kines seem promising for validation in prospective research, between pathogen directed antibiotic treatment and empirical
especially concerning viral aetiology. Future research should broad spectrum antibiotic treatment in patients with community
focus on validating our findings in a prospective unselected acquired pneumonia: a prospective randomised study. Thorax.
cohort of patients with CAP. 2005;60(8):672‐678.
6. Jennings LC, Anderson TP, Beynon KA, et al. Incidence and
characteristics of viral community‐acquired pneumonia in adults.
ACKNOWLEDGMENTS Thorax. 2008;63(1):42‐48.
7. Menendez R, Sahuquillo‐Arce JM, Reyes S, et al. Cytokine
The authors thank Mrs B. Dierdorp and T. Dekker (Amsterdam activation patterns and biomarkers are influenced by microor-
UMC, University of Amsterdam) for performing the assays ganisms in community‐acquired pneumonia. Chest. 2012;141(6):
and helping with the interpretation of results. The authors 1537‐1545.
thank Mr T. van der Ploeg for statistical assistance. None of 8. Farr BM, Kaiser DL, Harrison BD, Connolly CK. Prediction of mi-
the authors have conflicting interests. crobial aetiology at admission to hospital for pneumonia from the
presenting clinical features. British Thoracic Society Pneumonia
Research Subcommittee. Thorax. 1989;44(12):1031‐1035.
CONFLICT OF INTERESTS 9. Huijskens EG, Koopmans M, Palmen FM, van Erkel AJ, Mulder
PG, Rossen JW. The value of signs and symptoms in differentiat-
The authors declare that they have no conflicts of interest ing between bacterial, viral and mixed aetiology in patients with
with the contents of this article. community‐acquired pneumonia. J Med Microbiol. 2014;63(Pt
3):441‐452.
10. Ruiz M, Ewig S, Marcos MA, et al. Etiology of community‐
AUTHOR CONTRIBUTIONS acquired pneumonia: impact of age, comorbidity, and severity. Am
All authors contributed to the methodology of this article. J Respir Crit Care Med. 1999;160(2):397‐405.
11. Ruiz‐Gonzalez A, Falguera M, Vives M, Nogues A, Porcel JM,
EHB and RL analysed the samples. EHB and RD collected
Rubio‐Caballero M. Community‐acquired pneumonia: develop-
the samples, interpretated the data and wrote the manuscript. ment of a bedside predictive model and scoring system to identify
EHB, RD and WB have access to the full data. All authors the aetiology. Respir Med. 2000;94(5):505‐510.
revised and corrected the manuscript. 12. Schuetz P, Christ‐Crain M, Thomann R, et al. Effect of procalci-
tonin‐based guidelines vs standard guidelines on antibiotic use in
lower respiratory tract infections: the ProHOSP randomized con-
ETHICS trolled trial. JAMA. 2009;302(10):1059‐1066.
13. Christ‐Crain M, Jaccard‐Stolz D, Bingisser R, et al. Effect of pro-
The study was approved by the medical ethics committee.
calcitonin‐guided treatment on antibiotic use and outcome in lower
respiratory tract infections: cluster‐randomised, single‐blinded in-
ORCID tervention trial. Lancet. 2004;363(9409):600‐607.
14. Kruger S, Ewig S, Papassotiriou J, et al. Inflammatory parameters
Eduard H. Burgmeijer  https://orcid. predict etiologic patterns but do not allow for individual prediction
org/0000-0003-1618-3625 of etiology in patients with CAP: results from the German compe-
tence network CAPNETZ. Respir Res. 2009;10:65‐9921‐10‐65.
15. Roark CL, Simonian PL, Fontenot AP, Born WK, O'Brien RL.
R E F E R E NC E S gammadelta T cells: an important source of IL‐17. Curr Opin
Immunol. 2008;20(3):353‐357.
1. Wiersinga WJ, Bonten MJ, Boersma WG, et al. SWAB/ 16. Craig A, Mai J, Cai S, Jeyaseelan S. Neutrophil recruit-
NVALT (Dutch Working Party on Antibiotic Policy and ment to the lungs during bacterial pneumonia. Infect Immun.
Dutch Association of Chest Physicians) guidelines on the 2009;77(2):568‐575.
BURGMEIJER et al.   
|
   613

17. Martinez GJ, Nurieva RI, Yang XO, Dong C. Regulation and 30. Paats MS, Bergen IM, Hanselaar WE, et al. Local and systemic
function of proinflammatory TH17 cells. Ann N Y Acad Sci. cytokine profiles in nonsevere and severe community‐acquired
2008;1143:188‐211. pneumonia. Eur Respir J. 2013;41(6):1378‐1385.
18. Dehoux MS, Boutten A, Ostinelli J, et al. Compartmentalized cy- 31. Christ‐Crain M, Jaccard‐Stolz D, Bingisser R, et al. Effect of
tokine production within the human lung in unilateral pneumonia. procalcitonin‐guided treatment on antibiotic use and outcome in
Am J Respir Crit Care Med. 1994;150(3):710‐716. lower respiratory tract infections: cluster‐randomised, single‐
19. Hermesh T, Moltedo B, Moran TM, Lopez CB. Antiviral instruc- blinded intervention trial. Lancet. 2004;363(9409):600‐607.
tion of bone marrow leukocytes during respiratory viral infections. 32. Masia M, Gutierrez F, Padilla S, et al. Clinical characterisation
Cell Host Microbe. 2010;7(5):343‐353. of pneumonia caused by atypical pathogens combining clas-
20. Ravanetti L, Dijkhuis A, Sabogal Pineros YS, et al. An early in- sic and novel predictors. Clin Microbiol Infect. 2007;13(2):
nate response underlies severe influenza‐induced exacerbations 153‐161.
of asthma in a novel steroid‐insensitive and anti‐IL‐5‐responsive 33. Murdoch DR, Jennings LC, Bhat N, Anderson TP. Emerging ad-
mouse model. Allergy. 2016;72:727‐753. vances in rapid diagnostics of respiratory infections. Infect Dis Clin
21. Percopo CM, Dyer KD, Ochkur SI, et al. Activated mouse eo- North Am. 2010;24(3):791‐807.
sinophils protect against lethal respiratory virus infection. Blood. 34. Grumelli S, Corry DB, Song LZ, et al. An immune basis for lung
2014;123(5):743‐752. parenchymal destruction in chronic obstructive pulmonary disease
22. Hament JM, Kimpen JL, Fleer A, Wolfs TF. Respiratory viral in- and emphysema. PLoS Med. 2004;1(1):e8.
fection predisposing for bacterial disease: a concise review. FEMS 35. Monso E, Ruiz J, Rosell A, et al. Bacterial infection in chronic
Immunol Med Microbiol. 1999;26(3‐4):189‐195. obstructive pulmonary disease. A study of stable and exacerbated
23. Robinson KM, Choi SM, McHugh KJ, et al. Influenza A exacer- outpatients using the protected specimen brush. Am J Respir Crit
bates Staphylococcus aureus pneumonia by attenuating IL‐1beta Care Med. 1995;152(4 Pt 1):1316‐1320.
production in mice. J Immunol. 2013;191(10):5153‐5159. 36. Kellum JA, Kong L, Fink MP, et al. Understanding the inflam-
24. Robinson KM, Lee B, Scheller EV, et al. The role of IL‐27 in sus- matory cytokine response in pneumonia and sepsis: results of the
ceptibility to post‐influenza Staphylococcus aureus pneumonia. Genetic and Inflammatory Markers of Sepsis (GenIMS) Study.
Respir Res. 2015;16:10‐015‐0168‐8. Arch Intern Med. 2007;167(15):1655‐1663.
25. Stifter SA, Bhattacharyya N, Pillay R, et al. Functional interplay 37. Aldo P, Marusov G, Svancara D, David J, Mor G. Simple PlexTM:
between type I and II interferons is essential to limit influenza a novel multi‐analyte, automated microfluidic immunoassay plat-
a virus‐induced tissue inflammation. PLoS Pathog. 2016;12(1): form for the detection of human and mouse cytokines and chemok-
e1005378. ines. Am J Reprod Immunol. 2016;75(6):678‐693.
26. Endeman H, Meijvis SC, Rijkers GT, et al. Systemic cytokine
response in patients with community‐acquired pneumonia. Eur
Respir J. 2011;37(6):1431‐1438. SUPPORTING INFORMATION
27. Lieberman D, Livnat S, Schlaeffer F, Porath A, Horowitz S, Levy
R. IL‐1beta and IL‐6 in community‐acquired pneumonia: bacte- Additional supporting information may be found online in
remic pneumococcal pneumonia versus Mycoplasma pneumoniae the Supporting Information section at the end of the article. 
pneumonia. Infection. 1997;25(2):90‐94.
28. Zobel K, Martus P, Pletz MW, et al. Interleukin 6, lipopolysac-
charide‐binding protein and interleukin 10 in the prediction of risk How to cite this article: Burgmeijer EH, Duijkers R,
and etiologic patterns in patients with community‐acquired pneu-
Lutter R, Bonten MJM, Schweitzer VA, Boersma WG.
monia: results from the German competence network CAPNETZ.
Plasma cytokine profile on admission related to aetiology
BMC Pulm Med. 2012;20(12):6‐2466‐12‐6.
29. Andrijevic I, Matijasevic J, Andrijevic L, Kovacevic T, Zaric B. in community‐acquired pneumonia. Clin Respir J.
Interleukin‐6 and procalcitonin as biomarkers in mortality predic- 2019;13:605–613. https​://doi.org/10.1111/crj.13062​
tion of hospitalized patients with community acquired pneumonia.
Ann Thorac Med. 2014;9(3):162‐167.