Journal of the Pediatric Infectious Diseases Society

Original Article

Systematic Review and Meta-Analysis of Diagnostic

Biomarkers for Pediatric Pneumonia
Lourdes Cynthia Gunaratnam,1 Joan L. Robinson,1, and Michael T. Hawkes1,2,3,4,5,
Department of Pediatrics, University of Alberta, Edmonton, Alberta, Canada, 2Department of Medical Microbiology and Immunology, University of Alberta, Edmonton, Alberta,
1

Canada, 3School of Public Health, University of Alberta, Edmonton, Alberta, Canada, 4Distinguished Researcher, Stollery Science Lab, University of Alberta, Edmonton, Alberta,
Canada, and 5Member, Women and Children’s Research Institute, University of Alberta, Edmonton, Alberta, Canada

Background. Pneumonia causes significant morbidity and mortality in children worldwide, especially in resource-poor set-
tings. Accurate identification of bacterial etiology leads to timely antibiotic initiation, minimizing overuse, and development of
resistance. Host biomarkers may improve diagnostic sensitivity and specificity. We assessed the ability of biomarkers to correctly

Downloaded from https://academic.oup.com/jpids/article/10/9/891/6313111 by guest on 17 June 2025

identify bacterial pneumonia in children who present with respiratory distress.
Methods. A librarian-directed search was conducted of MEDLINE, EMBASE, CENTRAL, Global Health, the World Health
Organization International Clinical Trials Registry Platform, and ClinicalTrials.gov to May 2020 with no language restriction.
Included studies compared a diagnostic biomarker in children with bacterial pneumonia to those with nonbacterial respiratory
distress.
Results. There were 31 observational studies of 23 different biomarkers. C-reactive protein (CRP), procalcitonin (PCT), white
blood cell (WBC) count, and erythrocyte sedimentation rate (ESR) were the biomarkers with sufficient data for meta-analysis. Meta-
analysis revealed that CRP and PCT best differentiated bacterial from viral pneumonia with CRP summary AUROC (area under
the receiver operating characteristic curve) 0.71 (0.69-0.73), Youden index 53 mg/L, sensitivity 0.70 (0.68-0.78), and specificity 0.64
(0.58-0.68) and PCT summary AUROC 0.70 (0.67-0.74), Youden index 0.59 ng/mL, sensitivity 0.69 (0.65-0.77), and specificity 0.64
(0.60-0.68). WBC and ESR did not perform as well. Nineteen other inflammatory and immunologic biomarkers were identified in-
cluding CRP/mean platelet value, neutrophil/leukocyte ratio, interleukin 6, and interferon-alpha, with sensitivities from 60% to 85%
and specificities from 76% to 83%.
Conclusion. CRP and PCT performed better than WBC and ESR but had suboptimal sensitivity. Some less well-studied novel
biomarkers appear to have promise particularly in combination.
Key words. biomarker; pediatric; pneumonia; sensitivity; specificity.

Beyond the neonatal period, pneumonia continues to be the pneumonia, these regions continue to use the WHO revised
most common cause of mortality in children under the age of classification and treatment algorithm for diagnosis and treat-
5 worldwide, accounting for almost 13% of deaths [1]. Despite ment [3, 5]. These case management strategies have been shown
the reduction in childhood mortality achieved through the to have high sensitivity (62%-94%) but low specificity (16%-
Millennium Development Goal period (2000-2015), there were 20%) [6] for identifying bacterial etiology. As such, they have
138 million episodes of pediatric pneumonia in 2015, resulting been effective in decreasing pneumonia-related deaths by up to
in 0.8 million deaths [2]. Low-income countries in Sub-Saharan 36% [3, 7] but concerns of antibiotic overuse, depleting anti-
Africa and Southeast Asia have the highest burdens, accounting biotic stocks, and increasing microbial resistance are emerging
for more than three-quarters of all cases [1–3]. Most cases of [3, 7, 8]. Improving the specificity and sensitivity with which
pneumonia requiring hospitalization in these countries are bacterial pneumonia can be diagnosed is crucial to improving
caused by viral pathogens (61.4% as estimated by a recent large childhood pneumonia outcomes while maintaining antibiotic
study), with respiratory syncytial virus (RSV) being the most stewardship.
common [4]. However, due to the high morbidity of bacterial Although chest radiography (CXR) is frequently used to
diagnose pneumonia, it has important limitations including
poor interobserver agreement, decreased accessibility in
Received 26 December 2020; editorial decision 17 May 2021; accepted 26 May 2021; Published resource-poor settings, and most importantly for this anal-
online July 2, 2021.
Corresponding Author: Lourdes Cynthia Gunaratnam, MD, Department of Pediatrics,
ysis, an inability to distinguish bacterial pneumonia from
University of Alberta, Edmonton, Alberta, Canada. E-mail: gunaratn@ualberta.ca. viral pneumonia or atelectasis [9, 10]. Sputum cultures are
Journal of the Pediatric Infectious Diseases Society   2021;10(9):891–900 difficult to collect in children, can grow upper airway con-
© The Author(s) 2021. Published by Oxford University Press on behalf of The Journal of the
Pediatric Infectious Diseases Society. All rights reserved. For permissions, please e-mail:
taminants, and may not reflect lower respiratory tract dis-
journals.permissions@oup.com. ease [11–13]. Detection of a virus by molecular testing in an
DOI: 10.1093/jpids/piab043

Diagnostic Biomarkers in Pediatric Pneumonia • jpids 2021:10 (September) • 891

upper respiratory tract sample does not exclude coinfection Types of Outcome Measures
with bacteria [11, 13, 14]. Blood cultures are rarely positive Studies are needed to compare biomarker results in presumed
[15, 16]. Therefore, a sensitive and specific point-of-care test bacterial pneumonia to results in other causes of respiratory
would be of tremendous value. distress. The criteria to define bacterial pneumonia needed to
Given that bacteria and viruses evoke different host- be stated and rational and could include or exclude cases due to
pathogen interactions in pneumonia, quantifying biological Chlamydia pneumoniae and Mycoplasma pneumoniae.
molecules (ie, biomarkers) that reflect this differential response Two independent reviewers (L.C.G., J.L.R.) screened the ar-
offer diagnostic potential. Several biomarkers have been inves- ticles for inclusion, first by title and abstract and then by full
tigated to date, including (i) pathogen-related proteins and (ii) text. Disagreements were resolved by consensus.
host proteins, including markers related to inflammation, endo-
thelial function, and lung-specific proteins [17]. This systematic Data Extraction
review assessed the ability of biomarkers to correctly identify Two investigators (L.C.G., J.L.R.) extracted data pertaining to
bacterial pneumonia in children who present with respiratory study design, clinical setting, participants, target condition, ref-
distress. erence standards, index test, cutoff value, comparator test(s),

Downloaded from https://academic.oup.com/jpids/article/10/9/891/6313111 by guest on 17 June 2025

and outcomes (sensitivities and specificities) onto a standard-
METHODS
ized form. Disagreements were resolved by consensus between
the 2 reviewers.
The Preferred Reporting Items for Systematic Reviews and
Meta-Analyses guidelines were followed for conduct and re- Quality Appraisal
porting of the study (http://www.prisma-statement.org/) and Each study was assessed for risk of bias and applicability using
are reported in Supplementary Figure 1. QUADAS-2 by 2 investigators (L.C.G., J.L.R.). There is not a ref-
erence standard for the diagnosis of bacterial pneumonia. For
Search Criteria this item, studies that used plausible microbiological evidence
A librarian-directed search was conducted of the electronic in addition to CXR findings were considered to have a low risk
databases MEDLINE (1946 to May 2020), EMBASE (1974 to of bias whereas those that did not use both microbiological
May 2020), CENTRAL (inception to present), and Global and CXR findings were considered to have a high risk of bias.
Health (1920-2020) in May 2020 using the search terms Disagreements were resolved by consensus.
shown in Supplementary Figure 2. Ongoing trials were sought
at ClinicalTrails.gov and at the World Health Organization Statistical Analysis and Data Synthesis
International Clinical Trials Registry Platform (search per- Where sufficient data were available, summary receiver oper-
formed August 2018 but could not be updated in 2020 as the ating characteristic (ROC) curves were generated using a meta-
portal was temporarily not accessible due to COVID-19 traffic). analytic approach. To synthesize ROC curves from multiple
There were no language or format restrictions. Reference lists of diagnostic studies, we used the method of Martínez-Camblor
included articles were hand-searched. [18] implemented in the R Statistical environment [19] with
package nsROC [20]. In brief, this fully nonparametric ap-
Study Selection proach estimates an overall ROC curve using the information
Types of Studies of all cutoff points available in the selected original studies,
Any diagnostic research study with a control group and at least based on weighting each individual interpolated ROC curve
one comparison group. [18]. This method contrasts with meta-analytic techniques that
use a parametric approach, computing a bivariate estimation for
Types of Participants
the sensitivity and the specificity by using only one threshold
Children, birth to 18 years, were the focus. Studies with adults
per included study [18]. We used random-effects models to es-
were included if the results for children could be extracted.
timate the overall AUROC and statistically optimal sensitivity
Types of Interventions and specificity (Youden index). We used bootstrapping (1000
Studies of any biomarker, defined as biological molecules iterations) to estimate the 95% confidence intervals. We defined
found in any easily accessible bodily fluid which has the poten- the statistically optimal cutoff value of a biomarker as the value
tial to serve as a measurable indicator of bacterial pneumonia where the Youden index is maximized and used a parametric
were included. Studies looking solely at bronchoalveolar la- method to define this threshold [21].
vage fluid (BALF) were therefore excluded. The results of Where there was 1 outlier CRP cutoff value (0.065 mg/L),
the study had to include a cutoff value or an area under the a sensitivity analysis was performed by re-running the meta-
receiver operating characteristic curve (AUROC) for the analysis without this result to determine the effect of this value
biomarker. on the conclusions of the meta-analysis.

892 • jpids 2021:10 (September) • Gunaratnam et al

 To assess publication bias, we created funnel plots of the di- were observational; 4 were retrospective and 27 were prospec-
agnostic odds ratio for each of the biomarkers using package tive. Bacterial pneumonia was defined by one or more of the
meta [20] in the R Statistical environment [19]. following: compatible clinical findings (N = 4), compatible CXR
findings (N = 10), positive bacterial cultures from blood, other
sterile sites or the nasopharynx (N = 15), antigen or antibody
RESULTS
detection (N = 14), molecular detection of bacteria in blood,
Study Selection urine, or the nasopharynx (N = 7), and expert opinion (N = 1).
The search identified 2342 records of which 157 were screened Five studies analyzed only pneumococcal pneumonia, 2 ana-
by full text. Thirty-one studies (29 in English, 1 in Chinese, and lyzed only typical bacterial pneumonia (excluding cases sus-
1 in French) were eligible for inclusion (Figure 1). One other pected to be due to atypical pathogens such as M. pneumoniae),
study was excluded as we could not find a translator for the 13 combined typical and atypical bacterial pneumonia, and 11
Turkish language [22]. did not clarify whether atypical cases were included. The com-
parison condition of “other causes of respiratory distress” in-
Study Characteristics cluded viral pneumonia (N = 24) with 1 study specifying RSV

Downloaded from https://academic.oup.com/jpids/article/10/9/891/6313111 by guest on 17 June 2025

The characteristics of included studies are summarized in pneumonia, community-acquired pneumonia without consol-
Supplementary Table 1. Most were conducted in hospitals idation (N = 1), nonbacterial pneumonia based on CXR, low
(emergency department, admitting ward, or pediatric inten- white blood cell (WBC) count, negative cultures, and clinical
sive care unit [ICU]), one was conducted in an outpatient clinic findings (N = 2), bronchitis (N = 1), and viral or malarial illness
in Tanzania, and the setting of another study was unclear. All (N = 2). If viruses were detected in cases that otherwise met

Figure 1. Study selection.

Diagnostic Biomarkers in Pediatric Pneumonia • jpids 2021:10 (September) • 893

the criteria for bacterial pneumonia, 7 studies included the case Overall applicability concerns were low. However, 3
in the bacterial pneumonia group while 7 studies excluded the studies had concerns for their applicability based on pa-
case in their analysis; in the remaining 17 studies, the disposi- tient selection (one included only ICU patients, one focused
tion of such cases was unclear. on RSV alone as the comparator group, and one had a high
burden of human immunodeficiency virus coinfection). In
Study Quality 9 studies, there were concerns about the reference standard.
Results are summarized in Figure 2. Many of the studies lacked In 5 studies, only pneumococcus pneumonia was included
clarity in patient selection (N = 13). Four studied the same and in 1 study pneumococcus and atypical bacteria were
population. Seven studies had high risk of bias in the way the included; by failing to include all bacteria, these reference
index test (biomarker with cutoff) was interpreted; in 8 addi- standards were too narrow for the research question. In 1
tional studies, this was unclear. The 16 other studies that used study, C-reactive protein (CRP), a biomarker, was used to
either the AUROC or predetermined cutoff values had a low define the reference standard thereby removing it from the
risk of bias. Nine studies had a high risk of bias for the reference biomarker analysis. In 2 studies, the same applied to WBC.
standard as only one of CXR or microbiological testing (rather Finally, in one of these studies by Huang et al, there was also a

Downloaded from https://academic.oup.com/jpids/article/10/9/891/6313111 by guest on 17 June 2025

than both) was used to define it. In another 6 studies, it was un- high burden of malaria and it was not appropriately differen-
clear if the reference standard was interpreted without knowl- tiated from the reference standard. This negatively impacted
edge of the results. Thirteen studies had a high risk of bias with the reference standard applicability and the index test inter-
respect to flow and timing; most often these studies excluded pretation in this study.
patients with evidence for both viral and bacterial pneumonia,
no known pathogen, the “uncertain group,” “equivocal group,” Biomarker Identification
or complicated pneumonias. In an additional 5 studies, it was Tables 1-6 summarize sensitivities and specificities of the
unclear whether these patients were included. biomarkers at different cutoff values. The most commonly

Figure 2. Included studies and QUADAS-2 analysis results.

894 • jpids 2021:10 (September) • Gunaratnam et al

investigated biomarkers were CRP (Table 1), procalcitonin The results of the 19 other biomarkers are summarized
(PCT) (Table 2), WBC (Table 3), and erythrocyte sedi- in Table 5. These include immunological biomarkers (abso-
mentation rate (ESR) (Table 4). The CRP cutoffs studied lute neutrophil count [ANC], neutrophil/lymphocyte [N/L]
ranged from 8 mg/L to 200.15 mg/L; there was an outlier ratio), neutrophils, neutrophil to leukocyte ratio, lipocalin-2
of 0.065 mg/L which appears to be incorrectly reported and [Lpc-2], interferon-alpha [IFN-α], soluble triggering re-
may more likely be a cutoff of 65 mg/L. PCT cutoffs were ceptor expressed on myeloid cells-1 [sTREM-1]), inflam-
0.5-2 ng/mL with the exception of 2 studies, which used matory biomarkers (CRP/mean platelet value [CRP/MPV],
cutoffs of 0.18 ng/mL and 7 ng/mL. WBC cutoffs ranged interleukin 6 [IL-6], syndecan 4 proteoglycan, soluble glyco-
from 6 × 109/L to 18 × 109/L. ESR cutoffs ranged from 15 protein CHI3L1, interleukin 10 [IL-10]), vascular markers
to 120 mm/hr. CRP and PCT best differentiated bacterial (von Willebrand factor [vWF]), and others (midregional pro-
from viral pneumonia with CRP summary AUROC 0.71 atrial natriuretic peptide [MR-proANP], midregional pro-
(0.69-0.73), Youden index 53 mg/L, sensitivity 0.70 (0.68- adrenomedullin [MR-proADM], haptoglobin, tissue inhibitor
0.78), and specificity 0.64 (0.58-0.68) and PCT summary of metalloproteinase 1 [TIMP-1], tumor necrosis factor re-
AUROC 0.70 (0.67-0.74), Youden index 0.59 ng/mL, sensi- ceptor 2 [TNFR-2]). As a category, immunologic biomarkers

Downloaded from https://academic.oup.com/jpids/article/10/9/891/6313111 by guest on 17 June 2025

tivity 0.69 (0.65-0.77), and specificity 0.64 (0.60-0.68). In a performed best.
sensitivity analysis excluding the seemingly erroneous study Four studies looked at combinations of biomarkers (Table 6).
with CRP cutoff of 0.065 mg/L, the summary AUROC for TRAIL, IP-10, and CRP had a combined sensitivity of 95% and
CRP was 0.71 (0.68-0.75), showing that this study did not specificity of 94% while haptoglobin, TNFR-2 or IL-10, and
materially influence the conclusion of the meta-analysis. TIMP-1 combined had a sensitivity of 96% and specificity of

Table 1. Sensitivity and Specificity of CRP in Differentiating Bacterial Pneumonia From Nonbacterial Pneumonia

Study Total Sample Size Bacterial Pneumonia Nonbacterial Pneumonia Cutoff (mg/L) Sensitivity (%) Specificity (%)
Alcoba, 2017 142 50 92 >40 84 50
>80 74 70
Babu, 1989 60 30 30 >35 100 100
Bhuiyan, 2019 230 30 118 ≥40 83 60
≥60 75 77
≥72 75 84
≥100 67 87
Cevey-Macherel, 2009 85 52 33 >60 88 44
Elemraid, 2014 139 67 72 ≥20 32 78
>80 64 75
Erdman, 2015 124 30 94 >44 80 79
Esposito, 2016 90 74 16 >7.4 64 69
Esposito, 2016 346 235 111 >7.98 51 80
Gauchan, 2016 654 285 369 >36 62 91
Gendrel, 2002 72 43 29 >20 88 40
>60 70 52
Higdon, 2017 675 119 556 >37.1 77 82
Huang, 2014 103 54 49 >200.15 71 56
Korppi, 1993 67 30 37 >40 47 81
Marcus, 2008 51 36 15 >8 NR NR
Nohynek, 1995 84 54 30 >20 74 27
>40 41 63
>80 24 90
>120 13 93
Prat, 2003 85 51 34 >65 79 67
Toikka, 2000 108 68 40 >80 59 68
>150 31 88
Virkki, 2002 215 134 81 >80 52 72
≥20 78 33
>40 66 53
>120 36 85
Zhou, 2011 52 27 25 >0.065 93 28

Abbreviation: NR, not reported (no sensitivity/specificity reported, only AUROC).

Diagnostic Biomarkers in Pediatric Pneumonia • jpids 2021:10 (September) • 895

Table 2. Sensitivity and Specificity of PCT in Differentiating Bacterial Pneumonia From Nonbacterial Pneumonia

Study Total Sample Size Bacterial Pneumonia Nonbacterial Pneumonia Cutoff (ng/mL) Sensitivity (%) Specificity (%)
Alcoba, 2017 142 50 92 >0.5 81 46
>2 67 67
Cevey-Macherel, 2009 85 52 33 >5 72 58
Chen, 2016 192 96 96 >0.5 94 90
Don, 2007 41 18 23 >0.5 78 35
>1 56 61
>2 44 74
Esposito, 2016 90 74 16 >2 44 74
Erdman, 2015 124 30 94 >0.51 70 69
Esposito, 2016 346 235 111 >0.18 67 65
Gendrel, 2002 72 43 29 >0.5 95 60
>1 86 88
>2 63 96

Downloaded from https://academic.oup.com/jpids/article/10/9/891/6313111 by guest on 17 June 2025

Korppi, 2001 42 25 17 >0.5 56 71
>1 40 88
Korppi, 2003 86 57 29 >0.5 46 52
>1 24 90
Korppi, 2008 61 41 20 >1 59 64
Michelow, 2004 69 40 29 >0.75 68 69
Prat, 2003 85 51 34 >2 69 79
Toikka, 2000 108 68 40 >2 50 80
>7 19 98

86%; the other 2 combinations had low sensitivity. Alcoba et al DISCUSSION

also showed that combining CRP >40 mg/L with clinical find-
This systematic review and meta-analysis of 31 observational
ings and pneumococcal PCRs improved posttest probability of
studies showed that at least 23 biomarkers have been studied to
pneumonia with consolidation significantly from 35% to 77%.
differentiate bacterial pneumonia from nonbacterial causes of res-
piratory distress in children. Of these, CRP, PCT, WBC, and ESR
Meta-Analysis of ROC Curves for Selected Biomarkers have been studied most with CRP and PCT performing better
Summary ROC curves (Figure 3, Table 7) showed AUROC than WBC and ESR. Their sensitivity and specificity were about
(95% CI) for CRP 0.71 (0.69-0.73); PCT 0.70 (0.67-0.74); 70% and 65%, respectively, suggesting that they have inadequate
WBC 0.57 (0.55-0.60); and ESR 0.61 (0.53-0.66). Statistically diagnostic accuracy to be used on their own in clinical practice.
optimal cutoffs (based on the Youden index) were determined Literature reviews of biomarkers in pediatric pneumonia
(Table 7). Visual inspection of funnel plots (Supplementary have been published previously but have all been descriptive
Figure 3) did not reveal obvious asymmetry to suggest in nature without applying meta-analytic techniques for data
publication bias. synthesis. These studies summarize conflicting findings for the

Table 3. Sensitivity and Specificity of WBC in Differentiating Bacterial Pneumonia From Nonbacterial Pneumonia

Study Sample Size Target Condition (Bacterial Pneumonia) Comparison Condition (Nonbacterial Pneumonia) Cutoff (×109/L) Sensitivity (%) Specificity (%)
Bekdas, 2014 52 31 21 NRa 29 83
Elemraid, 2014 155 74 81 >15 58 63
Esposito, 2016 90 74 16 >10 74 39
Esposito, 2016 346 235 111 >13 46 61
Gendrel, 2002 72 43 29 >15 65 79
Korppi, 1993 67 30 37 >15 43 70
Nohynek, 1995 84 54 30 >12 52 47
>15 39 57
>18 22 80
Virkki, 2002 215 134 81 >15 48 53
Zhou, 2011 52 27 25 >6 96 28

Abbreviation: NR, not reported.

a
Used elevated WBC based on age-appropriate norms.

896 • jpids 2021:10 (September) • Gunaratnam et al

Table 4. Sensitivity and Specificity of ESR in Differentiating Bacterial to limit antibiotic overuse in pneumonia. The results are con-
Pneumonia From Nonbacterial Pneumonia flicting in the 2 pediatric studies to date [31, 32]. Esposito et al
Sample Bacterial Nonbacterial Cutoff Sensitivity Specificity found that PCT-guided antibiotic prescription significantly re-
Study Size Pneumonia Pneumonia (mm/hr) (%) (%)
duced antibiotic use and duration in Italy using a PCT cutoff of
Babu, 1989 60 30 30 >15 87 60 0.25 ng/mL. Using the same cutoff, Baer et al found a decreased
Korppi, 1993 67 30 37 >30 63 65
duration of antibiotics in Switzerland but no difference in anti-
Nohynek, 1995 84 54 30 >20 65 40
>30 41 73
biotic prescribing rate. They attributed the negative finding to
>40 31 87 low baseline prescribing rates and the low PCT cutoff, specu-
>120 13 93 lating that a higher cutoff would be more appropriate in the pe-
Virkki, 2002 215 134 81 >30 66 40 diatric population. It is of note that the optimal cutoff in our
study was higher at 0.59 ng/mL. One meta-analysis looking at
community-acquired bacterial pneumonia in adults showed
utility of various biomarkers and conclude that no single bio-
that PCT had moderate diagnostic accuracy with an AUROC
marker is accurate enough to predict etiology of pneumonia
of 0.771 (vs 0.70 in the current study) [29]. A subsequent

Downloaded from https://academic.oup.com/jpids/article/10/9/891/6313111 by guest on 17 June 2025

[17, 23–27]. A meta-analysis published 12 years ago evalu-
meta-analysis including 2408 adults and using the cutoff value
ated only CRP [28] and concluded that CRP values exceeding
of 0.5 mg/L showed unacceptably low sensitivity of 55% [33].
40-60 mg/L were able to weakly predict bacterial etiology (odds
Despite this, studies looking at clinical algorithms for adults
ratio 2.58; CI: 1.20-5.55). This cutoff is in keeping with the op-
incorporating point-of-care testing for PCT are underway
timal cutoff we identified (53 mg/L). A 2019 meta-analysis of
NCT03191071 [34], NCT03982667 [35]. Furthermore, adult
biomarkers to identify community-acquired bacterial pneu-
studies have used PCT levels to guide initiation and cessation of
monia in adults reported that CRP had the best diagnostic ac-
antibiotics in lower respiratory infections with a significant re-
curacy with a summary AUROC of 0.802 (vs 0.71 in the current
duction in overall antibiotic use and without adverse effects [36,
study) [29].
37]. It has been suggested that a “2 cut-point strategy” should be
A recent meta-analysis looked only at PCT as a diagnostic
studied, allowing for an indeterminate zone between PCT levels
biomarker in pediatric pneumonia and reported almost iden-
where antibiotics clearly are or are not indicated [38].
tical findings to ours [30]. There is great interest in using PCT
Of the other 19 biomarkers identified in our literature search,
the immunologic and inflammatory biomarkers performed
Table 5. Sensitivity and Specificity of Individual Biomarkers best, especially CRP/MPV, N/L ratio, IL-6, and IFN-α, with
Differentiating Bacterial Pneumonia From Nonbacterial Pneumonia sensitivities ranging between 60% and 85% and specificities be-
Study Sample Size Biomarker Cutoff Sensitivity (%) Specificity (%) tween 76% and 83%. Biomarkers identified by Huang et al ini-
Elemraid, 2014 152 Neutrophils >10 × 10 /L
9
59 72
tially showed great promise (Table 5). However, these results are
Esposito, 2016 346 Neutrophils >61% 64 54 most likely due to population bias in this study which had a high
Babu, 1989 60 ANC >10 000/mm3 40 100 burden of malaria. As a result, red blood cell biomarkers such
Nguyen, 2020 434 ANC ≥5 × 109/L 83 49 as haptoglobin and vWF are more likely to be differentiating
Nguyen, 2020 434 ANC ≥10 × 109/L 48 88
malaria from bacterial infection. This explains why the findings
Bekdas, 2014 52 N/L >1.7 74 76
could not be replicated in other populations as demonstrated by
Gauchan, 2016 654 N/L >1.28 46 64
Bekdas, 2014 52 CRP/MPV >2.6 81 81 the very different sensitivities and specificities that Lpc-2 had in
Gendrel, 2002 72 IL-6 >100 60 83 this study compared to the Esposito one (Table 5).
Toikka, 2000 108 IL-6 >40 34 83 Several novel putative biomarker “hits” which have not yet
Toikka, 2000 108 IL-6 >100 18 98 been validated were not included in the current review as cutoff
Esposito, 2016 90 Lpc-2 >1633 58 50
values are yet to be studied. Examples of approaches used for
Huang, 2014 103 Lpc-2 >130.1 84 86
host-based biomarker discovery include lectin microarray and
Erdman, 2015 124 CHI3L1 >57 ng/mL 93 63
Esposito, 2016 90 SYN4 >7.25 31 86 mass spectroscopy [39], RNA biosignatures induced by bacteria
Esposito, 2016 346 sTREM-1 >69 32 74 in a host [40], metabolomics, genomics, microbiomics, and
Esposito, 2016 346 MR-proANP <59 76 33 proteomics [41]. These technologies will undoubtedly identify
Esposito, 2016 346 MR-proADM >0.32 78 36 more biomarkers but as these biomarkers are identified, clear
Gendrel, 2002 72 IFN-α <2 85 83
diagnostic thresholds should be determined, and their diag-
Huang, 2014 103 vWF >1139 82 74
Huang, 2014 103 Haptoglobin >1.2 ng/mL 93 99
nostic accuracy was investigated.
Two studies that combined biomarkers (TRAIL, IP-10, and
Abbreviations: ANC, absolute neutrophil count; CRP/MPV, C-reactive protein/mean platelet volume; IFN-α,
interferon-alpha (U/mL); IL-6, interleukin 6 (pg/mL); Lpc-2, lipocalin 2 (ng/mL); MR-proADM, midregional pro- CRP and haptoglobin, TNFR-2 or IL-10, and TIMP-1) reported
adrenomedullin (nmol/L); MR-proANP, midregional pro-atrial natriuretic peptide (pmol/L); N/L, neutrophil to about 95% sensitivity and specificity (Table 6). These combin-
lymphocyte ratio; sTREM-1, serum soluble triggering expressed receptor on myeloid cells 1 (pg/mL); SYN4,
syndecan 4 proteoglycan (pg/mL); vWF, von Willebrand factor (mU/mL). ations clearly should be further studied, keeping in mind that

Diagnostic Biomarkers in Pediatric Pneumonia • jpids 2021:10 (September) • 897

Table 6. Studies Showing Sensitivity and Specificity of Combinations of Biomarkers to Differentiate Bacterial Pneumonia From Nonbacterial
Pneumonia

Study Sample Size Biomarkers and Cutoff Sensitivity (%) Specificity (%)
Don, 2009 39 Combo: CRP > 101 mg/L and WBC > 15.34 × 109/L and PCT > 1.0 ng/mL and ESR > 65 mm/h 35 87
Combo: CRP > 202 mg/L and WBC > 15.34 × 109/L and PCT > 1.0 ng/mL and ESR > 65 mm/h 21 100
Combo: CRP > 101 mg/L and WBC > 22 × 109/L and PCT > 1.0 ng/mL and ESR > 65 mm/h 36 100
Combo: CRP > 101 mg/L and WBC > 15.34 × 109/L and PCT > 18.0 ng/mL and ESR > 65 mm/h 29 93
Combo: CRP > 101 mg/L and WBC > 15.34 × 109/L and PCT > 1.0 ng/mL and ESR > 90 mm/h 14 100
Korppi, 2004 55 Any one of the following: 61 65
CRP > 80 mg/L, WBC > 17 × 109/L, PCT > 0.84 µg/mL, or ESR > 63 mm/hr
Mastboim, 2019 114 Assay combining TRAIL, IP-10, and CRP. Cutoff value <35 was diagnostic of viral; cutoff >65 was 95 94
diagnostic of bacterial; 35-65 was determined to be equivocal and not included.
Valim, 2016 80 Classification tree including haptoglobin, TNFR-2 or IL-10, and TIMP-1 96 86

Abbreviations: CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; IL-10, interleukin 10; IP-10, interferon-gamma–induced protein 10; PCT, procalcitonin; TIMP-1, tissue inhibitor of metalloproteinase 1; TNFR-2, tumor
necrosis factor receptor 2; TRAIL, tumor necrosis factor-related apoptosis-inducing ligand; WBC, white blood cells.

Downloaded from https://academic.oup.com/jpids/article/10/9/891/6313111 by guest on 17 June 2025

Figure 3. Summary of receiver operating characteristic (ROC) curves for studies of diagnostic biomarkers of pediatric pneumonia. (A) C-reactive protein
(CRP) with AUROC 0.71 (0.69-0.73). (B) Procalcitonin (PCT) with AUROC 0.70 (0.67-0.74). (C) Total white blood cell (WBC) count with AUROC 0.57 (0.55-0.60). (D)
Erythrocyte sedimentation rate (ESR) with AUROC 0.61 (0.53-0.66).

898 • jpids 2021:10 (September) • Gunaratnam et al

Table 7. Summary of Receiver Operating Characteristics Meta-Analysis study found that lower levels of CRP, WBC, ANC, and PCT
Sensitivity Specificity Youden were associated with M. pneumoniae infections vs typical bacte-
AUROC (Youden Index) (Youden Index) Index rial pneumonia, highlighting that atypical infections may have
CRP 0.71 (0.69-0.73) 0.70 (0.68-0.78) 0.64 (0.58-0.68) 53 mg/L biomarker effects more similar to viral pneumonias and quite
PCT 0.70 (0.67-0.74) 0.69 (0.65-0.77) 0.64 (0.60-0.68) 0.59 ng/mL different from typical bacterial pneumonias [43].
WBC 0.57 (0.55-0.60) 0.63 (0.37-0.77) 0.48 (0.36-0.73) 13 × 109
Another limitation of the current review is that many of
ESR 0.61 (0.53-0.66) 0.60 (0.40-0.65) 0.61 (0.50-0.69) 31 mm/hr
the studies were conducted prior to the widespread use of
Abbreviations: AUROC, area under the receiver operating characteristic curve; CRP, C-reactive protein; ESR,
erythrocyte sedimentation rate; PCT, procalcitonin; WBC, white blood cells.
pneumococcal vaccines. The studied biomarkers may per-
form differently in a population with a lower prevalence of
pneumococcal pneumonia. As well, the method used for
assays for some of these biomarkers are only available to re- summary curves has 2 limitations: first, different studies were
search laboratories. Mastboim et al excluded “equivocal test used to generate different curves for different biomarkers, as a
results” from the sensitivity analysis, begging the question result, we cannot compare the diagnostic accuracy of one bi-
of how including this group might have changed the results. omarker to another using the area under the curves; second,

Downloaded from https://academic.oup.com/jpids/article/10/9/891/6313111 by guest on 17 June 2025

Valim et al used a step-by-step diagnostic algorithm, starting summary ROC curves are only approximations of the true
with more sensitive biomarkers first and combining them with ROC since the meta-analytic method of Martinez-Camblor
more specific biomarkers next. In this way, they were able to et al [18] uses a finite number of discrete data points rather
achieve high diagnostic accuracy (Table 6), suggesting another than a continuous predictor.
tactic for future studies combining biomarkers. Another study In conclusion, CRP and PCT are more sensitive than WBC
by Alcoba et al looked at individual biomarkers, but also com- or ESR for the diagnosis of bacterial pneumonia but are not
bined biomarkers, PCR, and clinical signs (data not shown) sufficiently sensitive to use in isolation. Future endeavors for
to improve diagnosis of pneumonia with consolidation (from identifying diagnostic biomarkers should focus on reproducing
35% pretest probability to 77% posttest probability) and com- data with less well-studied immunologic biomarkers (CRP/
plicated bacterial pneumonia (from 20% to 86%). Though MPV, N/L ratio, IL-6, and IFN-α). Novel biomarkers that are
data are still scarce, combining biomarkers with each other or being discovered using unbiased approaches deserve the same
with clinical features in algorithms appears to hold diagnostic rigor. Currently, there is no biomarker that should be used as a
promise. stand-alone test, but they show promise in combination or as an
An important strength of our study was the comprehen- adjunct to clinical signs in diagnostic algorithms.
sive search strategy and detailed inclusion and exclusion cri-
teria. In particular, we were able to capture a large number of
studies but focus more closely on relevant diagnostic studies by Supplementary Data
necessitating cutoff values for biomarkers. Supplementary materials are available at the Journal of the Pediatric
Infectious Diseases Society online.
A key limitation to our study was the lack of a gold standard
for bacterial pneumonia. As clearly identified previously [4],
there is no diagnostic finding that comes close to conferring Notes
100% sensitivity and 100% specificity. In the current review, Acknowledgments. We would like to thank Meghan Sebastianski, Robin
Featherstone, and Dagmara Chojeki for their work creating and running
only one study [42] used the “silver standard” of positive cul-
the search strategy. We would also like to thank Bonita Lee, MD, Henrique
ture from sterile sites described by O’Brien et al [4]. The other De Sa Ellwanger, MD, and Sophia Yip, MD for their work translating non-
13 studies marked as “low risk of bias” used some combination English articles. Thanks to Ben Vandermeer his contribution to the statis-
of microbiological testing and CXR findings. The remainder tical analysis.
Financial support. This work was supported by Women & Children’s
had high or unclear risk of bias due to the way they chose to Health Research Institute Resident/Fellow Trainee Research Grant and
define the reference standard. There was also significant varia- Strategy for Patient-Oriented Research Support Unit.
bility as to whether atypical bacterial infections were included Potential conflicts of interest. All authors: No reported conflicts. All au-
thors have submitted the ICMJE Form for Potential Conflicts of Interest.
within the definition of bacterial pneumonia. Thirteen studies
Conflicts that the editors consider relevant to the content of the manuscript
included both atypical and typical bacterial infections and only have been disclosed.
one of these studies separated out the outcomes, excluding atyp-
ical bacteria [42]. In other studies, it was not clarified whether
References
atypical pneumonias were considered to be bacterial. This lim-
1. Liu L, Oza S, Hogan D, et al. Global, regional, and national causes of under-5
itation makes it difficult to compare individual study results to mortality in 2000-15: an updated systematic analysis with implications for the
each other but also adds significant heterogeneity to the meta- Sustainable Development Goals. Lancet 2016; 388:3027–35.
2. McAllister DA, Liu L, Shi T, et al. Global, regional, and national estimates of pneu-
analysis. Rectifying these discrepancies for future diagnostic monia morbidity and mortality in children younger than 5 years between 2000
studies related to bacterial pneumonia is crucial. One recent and 2015: a systematic analysis. Lancet Glob Health 2019; 7(1):e47–57.

Diagnostic Biomarkers in Pediatric Pneumonia • jpids 2021:10 (September) • 899

 3. Wardlaw TM, Johansson EW, Hodge M. Pneumonia: The Forgotten Killer of 23. Giulia B, Luisa A, Concetta S, et al. Procalcitonin and community-acquired pneu-
Children. New York, NY: The United Nations Children’s Fund (UNICEF)/World monia (CAP) in children. Clin Chim Acta 2015; 451:215–8.
Health Organization (WHO); 2006:1–40. 24. Zar HJ, Andronikou S, Nicol MP. Advances in the diagnosis of pneumonia in chil-
4. O’Brien KL, Baggett HC, Brooks WA, et al. Causes of severe pneumonia requiring dren. BMJ 2017; 358:j2739.
hospital admission in children without HIV infection from Africa and Asia: the 25. Esposito S, Principi N. Defining the aetiology of paediatric community-acquired
PERCH multi-country case-control study. Lancet 2019; 394(10200):757–79. pneumonia: an unsolved problem. Expert Rev Respir Med 2019; 13:153–61.
5. WHO. Revised WHO Classification and Treatment of Pneumonia in Children 26. Savvateeva EN, Rubina AY, Gryadunov DA. Biomarkers of community-acquired
at Health Facilities: Evidence Summaries. Geneva, Switzerland: World Health pneumonia: a key to disease diagnosis and management. Biomed Res Int 2019;
Organization; 2014:1–34. 2019:1701276.
6. Cardoso MR, Nascimento-Carvalho CM, Ferrero F, et al. Adding fever to WHO 27. Thomas J, Pociute A, Kevalas R, et al. Blood biomarkers differentiating viral
criteria for diagnosing pneumonia enhances the ability to identify pneumonia versus bacterial pneumonia aetiology: a literature review. Ital J Pediatr 2020;
cases among wheezing children. Arch Dis Child 2011; 96:58–61. 46:4.
7. Sazawal S, Black RE; Pneumonia Case Management Trials Group. Effect of 28. Flood RG, Badik J, Aronoff SC. The utility of serum C-reactive protein in
pneumonia case management on mortality in neonates, infants, and preschool differentiating bacterial from nonbacterial pneumonia in children: a meta-
children: a meta-analysis of community-based trials. Lancet Infect Dis 2003; analysis of 1230 children. Pediatr Infect Dis J 2008; 27:95–9.
3:547–56. 29. Ebell MH, Bentivegna M, Cai X, et al. Accuracy of biomarkers for the diagnosis of
8. Bhutta ZA, Gonzales R, Corbett KK, Wirtz V, Dreser A. Drug resistant infections adult community-acquired pneumonia: a meta-analysis. Acad Emerg Med 2020;
in poor countries. BMJ 2008; 336(7650):948–9. 27:195–206.
9. Andronikou S, McHugh K, Abdurahman N, et al. Paediatric radiology seen from 30. Tsou PY, Rafael J, Ma YK, et al. Diagnostic accuracy of procalcitonin for bacterial

Downloaded from https://academic.oup.com/jpids/article/10/9/891/6313111 by guest on 17 June 2025

Africa. Part I: providing diagnostic imaging to a young population. Pediatr Radiol pneumonia in children - a systematic review and meta-analysis. Infect Dis (Lond)
2011; 41:811–25. 2020; 52:683–97.
10. Elemraid MA, Muller M, Spencer DA, et al.; North East of England Paediatric 31. Esposito S, Tagliabue C, Picciolli I, et al. Procalcitonin measurements for guiding
Respiratory Infection Study Group. Accuracy of the interpretation of chest antibiotic treatment in pediatric pneumonia. Respir Med 2011; 105:1939–45.
radiographs for the diagnosis of paediatric pneumonia. PLoS One 2014; 32. Baer G, Baumann P, Buettcher M, et al. Procalcitonin guidance to reduce anti-
9:e106051. biotic treatment of lower respiratory tract infection in children and adolescents
11. Hammitt LL, Murdoch DR, Scott JA, et al.; Pneumonia Methods Working Group. (ProPAED): a randomized controlled trial. PLoS One 2013; 8:e68419.
Specimen collection for the diagnosis of pediatric pneumonia. Clin Infect Dis 33. Kamat IS, Ramachandran V, Eswaran H, et al. Procalcitonin to distinguish viral
2012; 54(Suppl 2):S132–9. from bacterial pneumonia: a systematic review and meta-analysis. Clin Infect Dis
12. Murdoch DR, Morpeth SC, Hammitt LL, et al.; PERCH Study Group. Microscopic 2020; 70:538–42.
analysis and quality assessment of induced sputum from children with pneu- 34. Noemie B-B. An algorithm to decide on antibiotic prescription in lower respira-
monia in the PERCH study. Clin Infect Dis 2017; 64:271–9. tory tract infection in primary care. ClinicalTrials.gov. 2017.
13. Thea DM, Seidenberg P, Park DE, et al.; PERCH Study Group. Limited utility 35. Mazlan MZ. Efficacy and safety of point-of-care procalcitonin test to reduce anti-
of polymerase chain reaction in induced sputum specimens for determining the biotic exposure in VAP. ClinicalTrials.gov. 2019.
causes of childhood pneumonia in resource-poor settings: findings from the 36. Christ-Crain M, Jaccard-Stolz D, Bingisser R, et al. Effect of procalcitonin-
pneumonia etiology research for child health (PERCH) study. Clin Infect Dis guided treatment on antibiotic use and outcome in lower respiratory tract in-
2017; 64:289–300. fections: cluster-randomised, single-blinded intervention trial. Lancet 2004;
14. Michelow IC, Olsen K, Lozano J, et al. Epidemiology and clinical characteristics 363:600–7.
of community-acquired pneumonia in hospitalized children. Pediatrics 2004; 37. Schuetz P, Christ-Crain M, Thomann R, et al.; ProHOSP Study Group. Effect of
113:701–7. procalcitonin-based guidelines vs standard guidelines on antibiotic use in lower
15. Hickey RW, Bowman MJ, Smith GA. Utility of blood cultures in pediatric patients respiratory tract infections: the ProHOSP randomized controlled trial. JAMA
found to have pneumonia in the emergency department. Ann Emerg Med 1996; 2009; 302:1059–66.
27:721–5. 38. Peacock WF, Rafique Z. Procalcitonin cut-point strategies. Clin Infect Dis 2020;
16. Shah SS, Alpern ER, Zwerling L, et al. Risk of bacteremia in young children with 71:245–6.
pneumonia treated as outpatients. Arch Pediatr Adolesc Med 2003; 157:389–92. 39. Yang L, Yang Z, Cheng L, et al. Lectin microarray combined with mass spec-
17. Florin TA, Ambroggio L. Biomarkers for community-acquired pneumonia in the trometry identifies haptoglobin-related protein (HPR) as a potential serologic
emergency department. Curr Infect Dis Rep 2014; 16:451. biomarker for separating nonbacterial pneumonia from bacterial pneumonia in
18. Martínez-Camblor P. Fully non-parametric receiver operating characteristic childhood. Proteomics Clin Appl 2018; 12:e1800030.
curve estimation for random-effects meta-analysis. Stat Methods Med Res 2017; 40. Lydon EC, Henao R, Burke TW, et al. Validation of a host response test to distin-
26(1):5–20. guish bacterial and viral respiratory infection. EBioMedicine 2019; 48:453–61.
19. R Core Team. R: a language and environment for statistical computing. https:// 41. Karakioulaki M, Stolz D. Biomarkers in pneumonia-beyond procalcitonin. Int J
www.R-project.org/. Accessed 3 June 2020. Mol Sci 2019; 20(8):1–18.
20. Pérez-Fernández S, Martínez-Camblor P, Filzmoser P, Corral N. nsROC: an R 42. Higdon MM, Le T, O’Brien KL, et al.; PERCH Study Group. Association of
package for non-standard ROC curve analysis. R J 2018; 10(2):55–77. C-reactive protein with bacterial and respiratory syncytial virus-associated pneu-
21. Steinhauser S. Determining optimal cut-offs in the meta-analysis of diagnostic monia among children aged <5 years in the PERCH study. Clin Infect Dis 2017;
test accuracy studies. 2018. https://d-nb.info/1120020735/34. Accessed 3 June 64:378–86.
2020. 43. Meyer Sauteur PM, Krautter S, Ambroggio L, et al. Improved diagnostics
22. Poyrazoðlu MH, Per H, Öztürk M, Bingöl N, Üzüm KM. Çocukluk çaðý help to identify clinical features and biomarkers that predict Mycoplasma
pnömonilerinde serum prokalsitonin düzeyleri. Çocuk Saðlýðý ve Hastalýklarý pneumoniae community-acquired pneumonia in children. Clin Infect Dis
Dergisi 2002; 46:169–76. 2020; 71:1645–54.

900 • jpids 2021:10 (September) • Gunaratnam et al