Abstract
Introduction:
Management of children ≤36 months of age presenting with fever without source remains a challenge because the underlying aetiologies may range from self-limiting viral infections to serious bacterial infections (SBIs) including bacteraemia, urinary tract infection (UTI), pneumonia, bacterial meningitis, osteomyelitis, or septic arthritis. This systematic review was conducted to determine the thresholds at which C-reactive protein (CRP), procalcitonin (PCT), and interleukin-6 (IL-6) can predict SBIs in this population.
Methods:
We systematically searched electronic databases such as MEDLINE, Cochrane, CINAHL, and Web of Science for studies that evaluated the diagnostic accuracies of CRP, PCT, and IL-6 in detecting SBIs in children ≤36 months of age presenting with fever without source, during the period between November 2013 and November 2023. Area under the summary receiver operating curve (SROC) was calculated by the Rutter and Gatsonis method. I2 was used to quantify study heterogeneity. All tests were two-sided, and a p-value <0.05 was considered statistically significant. This review was registered with PROSPERO, CRD42023439093.
Results:
Datasets from 37 studies were included. A CRP cut-off of 10–20 mg/L had the highest pooled sensitivity of 0.75 (95% CI: 0.54–0.89), while a CRP cut-off of >40 mg/L had the highest pooled specificity 0.92 (95% CI: 0.87–0.95). A PCT cut-off of <0.5 ng/mL had the highest pooled sensitivity of 0.7812 (95% CI: 0.59–0.90) but the lowest pooled specificity of 0.69 (95% CI: 0.54–0.81). Based on receiver operating curve (ROC) analysis, a CRP cut-off between 10 and 20 mg/L and a PCT cut-off of <0.5 ng/mL showed the best diagnostic performance with a pooled AUC of 0.84 (95% CI: 0.79–0.90) and 0.816 (95% CI: 0.727–0.928), respectively. Only 1 study for IL-6 reported that a threshold of 20 pg/dL had a sensitivity and specificity of 79.1% and 91.6%, respectively.
Conclusion:
A PCT cut-off of 0.5 ng/mL and a CRP cut-off of 10–20 mg/L display the best performance in identifying SBIs in children ≤36 months of age with fever without source.
Systematic Review Registration:
https://www.crd.york.ac.uk/PROSPERO/view/CRD42023439093, identifier CRD42023439093.
Introduction
Fever is the most common reason for paediatric emergency department visits (). Approximately 20% of febrile young children have undifferentiated fever without an apparent source of infection (). Although a majority of these patients have viral infections, a subset may have serious bacterial infections (SBIs) such as lobar pneumonia, urinary tract infections, bacteremia, and bacterial meningitis, which require appropriate management with antibiotics and hospitalization (). While untreated SBIs can lead to significant morbidity and mortality, over-treatment of self-limiting viral illnesses can lead to antibiotic resistance and higher healthcare utilization and costs (, ).
Clinical evaluation is challenging in children younger than 36 months, as the features distinguishing self-limiting viral infections from SBIs are often subtle and can be easily missed. To address this, many studies have been performed to evaluate how accurately adjunct laboratory markers can help predict SBIs in this group of patients. White blood cell (WBC), C-reactive protein (CRP), and procalcitonin (PCT) have been widely reported as biomarkers of inflammation (, ). Indeed, these biomarkers are incorporated in clinical predictive algorithms to predict risks of SBIs among young children presenting with fever without source. More recently, newer evidence has highlighted the emerging role of cytokines such as interleukin-6 (IL-6) in predicting pediatric sepsis ().
Interestingly, these biomarkers exhibit different kinetic profiles. CRP is an acute-phase reactant produced by the liver in response to tissue damage and infections, and its level can be elevated within 24 h of a trigger, peaks within 48 h, and has a half-life of 24 h (–). On the other hand, PCT levels can increase rapidly within 6–12 h in response to bacterial endotoxins, PCT peaks within 12 h, and it has a half-life of 24 h (). As such, PCT has been regarded as a better biomarker to identify early bacterial infections compared with CRP (). IL-6 is an inflammatory cytokine produced by monocytes and macrophages early in the inflammatory cascade, reaching its peak within 24 h but becoming undetectable within 24 h (, ).
Although the utility of these biomarkers has been widely reported and well established in the prediction of SBIs, the exact thresholds used to determine the likely presence of SBIs vary considerably among different studies (–). Therefore, in this study, we aimed to perform a systematic review and meta-analysis with the primary objective of determining the optimal thresholds at which CRP, PCT, and IL-6 can perform to distinguish the presence of SBIs in children younger than 36 months presenting with fever without source.
Materials and methods
We adhered to the methods and procedures of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Diagnostic Test Accuracy (PRISMA-DTA) checklist for the reporting of this systematic review and meta-analysis (). The study protocol has been published (). This review was registered with the International Prospective Register of Systematic Reviews with registration number CRD42023439093.
Eligibility criteria
We included studies conducted between November 2013 and November 2023 that reported specific thresholds and the corresponding diagnostic accuracies of CRP, PCT, and IL-6 in detecting SBIs in children younger than 36 months presenting with fever without source. Randomized controlled trials (RCTs) and non-randomized studies including non-RCTs, controlled before-and-after studies, and cohort studies were also included. We excluded case series, case-control studies, systematic reviews, meta-analyses, conference abstracts, and studies not reported in English. We also excluded studies that reported on both paediatric and adult patients but did not separately report on paediatric patients, studies with small population sizes (N < 50) (), and studies that examined high-risk populations, specifically infants with prematurity and/or with a history of immunodeficiencies and malignancies.
Information sources and search strategy
We performed a comprehensive search of the databases of MEDLINE, Cochrane Central Register of Controlled Trials, Cumulative Index to Nursing and Allied Health Literature (CINAHL), and Web of Science for pertinent studies. Additional references were identified by cross-checking bibliographies of full-text publications.
The search strategy was developed in collaboration with a librarian at our institution and is described in detail under Supplementary Information. The Medical Subject Headings (MeSH) terms “C-reactive protein,” “procalcitonin,” “interleukin-6,” “paediatrics,” “children,” “fever,” “pyrexia,” “infection,” and “sepsis” were used, and these terms were exploded where available. The search was conducted from 30 November 2023 to 1 December 2023. Duplicate results were removed.
Study selection
Screening was conducted by five independent authors (NS, LJ, SY, SG, and CS-L) using Covidence, a software that follows the PRISMA 2009 guidelines that facilitate screening of studies during the selection process (). Following initial screening of titles and abstracts, full-text articles of all relevant articles were retrieved and assessed based on study inclusion and exclusion criteria. Both stages of screening were performed by two independent reviewers, and any disagreements that followed were resolved by a third independent reviewer or by means of consensus.
Data extraction and synthesis
All relevant data were extracted using standardized data collection forms on Microsoft Excel and verified. We extracted the following data from each included article: study title, authors, publication date and year, type of publication, country, study design, sample size, inclusion and exclusion criteria, clinical setting, characteristics of study population (including age, gender, and comorbidities), and duration of fever at the time of diagnostic workup. We documented the need for hospital admission and duration of hospitalization, as well as antibiotic use. We obtained the method of sampling CRP, PCT, and IL-6, values and units of measurement for CRP, PCT, and IL-6, and diagnostic performance indicators such as true positives (TPs), true negatives (TNs), false positives (FPs), and false negatives (FNs). However, a study was excluded when the aforementioned parameters were not mentioned or could not be obtained from the receiver operating curve (ROC). We also extracted unadjusted and adjusted odds ratios (ORs) and the unadjusted area under the curve (AUC) with a 95% confidence interval (CI).
Definitions
Fever without source: temperature >38°C for less than 7 days without any signs or symptoms identifying the source of infection.
Early presentation: presentation of febrile children to the emergency department within 48 h of fever onset ().
SBI: presence of bacteraemia, urinary tract infection (UTI), pneumonia, bacterial meningitis, osteomyelitis, or septic arthritis (, ).
Bacteraemia: growth of a single bacterial pathogen in blood, excluding growth of bacteria considered a priori as contaminants, for example, coagulase-negative Staphylococcus ().
UTI: growth of single urine pathogen with at least 50,000 colony-forming units (CFUs)/mL from a catheterized urine specimen; or 10,000–50,000 CFUs/mL from a catheterized specimen with positive urinalysis (positive for leucocyte esterase, nitrite or pyuria at >5 white blood cells; or at least 100,000 CFUs/mL from urine collected via voided specimens) ().
Pneumonia: lobar consolidation diagnosed on chest radiography confirmed by a paediatric radiologist ().
Bacterial meningitis: cerebrospinal fluid leucocytes >5 cell/μL and positive bacterial culture (, ).
Osteomyelitis: positive bone scintigraphy.
Septic arthritis: positive bacterial culture of synovial fluid.
Risk-of-bias assessment
We assessed the validity of each study using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) assessment tool, which provides a standardized way of assessing the risk of bias and applicability of diagnostic test accuracy research across four domains: patient selection, index test, reference standard, and flow and timing (). Assessors are guided by signalling questions, which were adapted to this review, to reach a classification of “high,” “low,” or “unclear” for each subsection. The risk of bias or applicability was deemed to be “low risk” if all responses to signalling questions in that domain were affirmative; a classification of “high risk” was given if there were answers in the negative. If any question could not be answered because of a lack of information in the included study, the outcome for that subsection was “unclear.”
Statistical analysis
The extracted number of TPs, FPs, FNs, and TNs from each study was pooled to calculate sensitivity, specificity, positive likelihood ratio, negative likelihood ratio, and diagnostic OR (DOR) using bivariate random-effects models for diagnostic meta-analysis. The summary receiver operating characteristic curve (SROC) was plotted, and the area under the SROC was calculated by the Rutter and Gatsonis method (hierarchical regression approach) (). Based on available cut-offs from different studies, clinically meaningful categories were defined. For CRP, they were defined as <10, 10–<20, 20–<40, and ≥40 mg/L, while for procalcitonin, the cut-offs were <0.5 ng/mL, 0.5 ng/mL, and > 0.5–<2.0 mg/mL. A separate analysis was also done for CRP and procalcitonin to determine the best cut-off using Youden's index, with all the available cut-offs employed as continuous data.
Review Manager 5.4.1 (Cochrane Collaboration) analysis software was used to build the area under the SROC curve graphs by making use of different colours for different categories of CRP or procalcitonin cut-offs. Diagnostic power was considered good if the AUC was more than 0.8 and poor if it was less than 0.7. Publication bias was assessed using funnel plots, and Deek's asymmetry test was conducted to evaluate the publication bias for each category of CRP or procalcitonin cut-offs. A p-value of <0.10 indicated the presence of publication bias. All meta-analyses of unadjusted and adjusted ORs and the unadjusted AUC was performed using the DerSimonian Laird random effects model. Adjustments to ORs were based on the covariates identified in each individual study.
I2 statistic was used to quantify heterogeneity across studies, and an I2 statistic of 80% or more was considered to indicate considerable heterogeneity. However, owing to the broad review question resulting in an inherently heterogeneous population among included studies, we accepted I2 statistics up to 95% for inclusion in the meta-analysis (). All tests were two–sided, and p < 0.05 was considered statistically significant. All analyses were carried out using Meta-DiSc 2.0 (Hospital Ramon y Cajal and Universidad Complutense de Madrid, Madrid, Spain), Revman 5.4.1, and R software version 4.2.2 (2022-10-31 ucrt).
Results
Electronic database searches yielded a total of 2,299 publications with an additional 9 studies from manual searches for CRP, 965 publications with an additional 8 studies from manual searches for PCT, and 593 publications for IL-6 (Figures 1A–C). After exclusion of duplicates and screening of titles and abstracts of relevant publications, 66 studies were retrieved in full for review for CRP, 41 studies were retrieved in full for PCT, and 5 studies were retrieved for IL-6. Ultimately, datasets from 37 distinct studies were included in the meta-analysis, with 31 studies included in the meta-analysis for CRP and 25 studies included in the meta-analysis for PCT (Figures 1A–C). Only one study fulfilled the exclusion and inclusion criteria for IL-6.
Figure 1
The descriptive characteristics of the included studies, such as year of publication, study design and settings, number of eligible participants, and definitions of SBIs, are presented in Table 1. All studies were published in peer-reviewed journals. A majority (32 studies, 86.5%) of the studies were single-centre ones. Twenty-four studies (64.9%) had a retrospective study design. A majority (23 studies, 62.2%) included study participants up to 3 months of age. Fifteen studies (40.5%) defined SBIs as meningitis, bacteremia, UTI, and/or pneumonia, while eight studies (21.6%) defined SBIs as meningitis, bacteremia, and UTI. Seven studies (18.9%) included only the outcomes of bacteremia and/or meningitis. The reported rate of prevalence of SBIs ranged from 1.44% to 45.2%.
Table 1
| Citation | Country | Single or multi-centre | Study design | No. of participants | Age of study participants (months) | SBI prevalence (%) | Gender, female (%) | Cut-off values | Outcomes | |
|---|---|---|---|---|---|---|---|---|---|---|
| CRP (mg/L) | Procalcitonin (ng/mL) | |||||||||
| Sutiman et al. () | Singapore | Single | Prospective | 258 | 0–3 | 86 (33.3) | 109 (42.4) | 20 | 0.5, 1.7 | Meningitis, bacteremia, and UTI |
| Hiremath et al. () | India | Single | Explorative | 31 | 3–36 | 14 (45.2) | 12 (38.7) | 10 | 0.5 | Meningitis, bacteremia, UTI, and pneumonia |
| Nosrati et al. () | Israel | Single | Retrospective | 401 | 0–3 | 48 (12.0) | 177 (44.1) | 2, 4, 5, 10, 20, 30, 40 | N.A. | Meningitis, bacteremia, UTI, and pneumonia |
| Khoo et al. () | Singapore | Single | Retrospective | 659 | 0–3 | 161 (24.4) | 263 (39.9) | 20 | 0.5 | Meningitis, bacteremia, UTI, pneumonia, osteomyelitis, and septic arthritis |
| Manzano et al. () | Canada | Single | Prospective | 328 | 1.36 | 54 (16.0) | 163 (49.7) | 17.7 | 0.2 | Meningitis, bacteremia, UTI, pneumonia, osteomyelitis, and septic arthritis |
| Milcent et al. () | France | Multi | Prospective | 2,047 | 0.23–3 | 139 (6.8) | 829 (40.5) | 20, 40 | 0.3, 0.5, 2 | Meningitis, bacteremia, UTI, and pneumonia |
| Diaz et al. () | Spain | Single | Retrospective | 318 | 0–3 | 76 (23.9) | 128 (40.3) | 3 | 0.5, 2 | Meningitis and bacteremia |
| Lee et al. () | South Korea | Single | Retrospective | 336 | 1–3 | 38 (11.3) | 124 (36.9) | 0.5 | 0.5 | Meningitis, bacteremia, and UTI |
| Leroy et al. () | Switzerland | Multi | Retrospective | 877 | 0–36 | 211 (24.1) | 449 (51.2) | N.A. | 0.5, 2 | Meningitis, bacteremia, UTI, pneumonia, osteomyelitis, and septic arthritis |
| Chiu et al. () | Taiwan | Single | Retrospective | 1,231 | 0–3 | 49 (3.98) | 465 (37.8) | 25 | N.A. | Meningitis and bacteremia |
| D’Souza et al. () | India | Single | Retrospective | 240 | 0–24 | 44 (18.3) | 108 (45.0) | 7 | N.A. | Meningitis, bacteremia, UTI, and pneumonia |
| Galetto-Lacour et al. () | Switzerland | Single | Subpopulation of prospective database | 135 | 0–36 | 20 (14.8) | 55 (40.7) | 40 | 0.5 | Meningitis, bacteremia, UTI, pneumonia, and osteomyelitis |
| Hamiel et al. () | Israel | Single | Retrospective | 1,039 | 0.23–3 | 111 (10.6) | 411 (39.6) | 5, 20, 40, 80 | N.A. | Meningitis, bacteremia, and UTI |
| Gomez et al. () | Spain | Single | Subpopulation of prospective database | 1,527 | 0–3 | 22 (1.44) | 670 (43.9) | 20 | 0.5 | Meningitis, bacteremia, and UTI |
| Vujevic et al. () | Croatia | Single | Retrospective | 181 | 0–3 | 70 (38.7) | 71 (39.2) | 13.3 | N.A. | Meningitis, bacteremia, UTI, and pneumonia |
| Zarkesh et al. () | Iran | Single | Prospective | 195 | 0–3 | 29 (14.9) | 83 (42.6) | 10 | N.A. | Meningitis, bacteremia, UTI, and pneumonia |
| Polat et al. () | Turkey | Single | Retrospective | 127 | 1–36 | 27 (21.3) | 44 (34.6) | 9 | N.A. | Meningitis, bacteremia, UTI, and pneumonia |
| Rodriguez et al. () | Colombia | Single | Retrospective | 137 | 0–36 | 40 (29.2) | 78 (57.0) | 80 | N.A. | Meningitis, bacteremia, UTI, and pneumonia |
| Han et al. () | South Korea | Single | Retrospective | 199 | 1–3 | 68 (34.2) | - | 16.74, 20, 40 | 0.5, 0.6, 1.0 | Meningitis, bacteremia, and UTI |
| Markic et al. () | Croatia | Single | Retrospective | 135 | 0–6 | 62 (45.9) | 64 (47.4) | 40 | 0.5 | Meningitis, bacteremia, UTI, and pneumonia |
| Moldovan et al. () | Romania | Single | Prospective | 90 | 0.25–12 | 19 (21.1) | 40 (44.4) | 20, 40, 80 | 0.5, 2, 10 | Meningitis, bacteremia, UTI, and pneumonia |
| Park et al. () | South Korea | Single | Retrospective | 317 | 0–3 | 61 (19.2) | 123 (38.8) | 2 | 0.3 | Meningitis, bacteremia, UTI, and pneumonia |
| Parada et al. () | Spain | Single | Retrospective | 220 | 0–3 | 85 (38.6) | 32 (40.0) | 1, 2.75, 4, 10 | 0.26, 0.5, 1 | Meningitis, bacteremia, UTI, and pneumonia |
| Li et al. () | Taiwan | Single | Retrospective | 131 | 0–1 | 38 (29.0) | 51 (39.0) | 2.5 | N.A. | Meningitis, bacteremia, UTI, and pneumonia |
| Hernandez-Bou et al. (55) | Spain | Single | Prospective | 283 | 1–24 | 34 (12.0) | 120 (42.4) | 10 | 0.12 | Meningitis and bacteremia |
| Kim et al. (56) | South Korea | Single | Retrospective | 133 | 3–36 | 43 (32.0) | 62 (46.6) | 78 | N.A. | Meningitis, bacteremia, UTI, pneumonia, osteomyelitis, and septic arthritis |
| Lacroix et al. (57) | Switzerland | Single | Prospective | 241 | 0–36 | 41 (17.0) | 93 (39.0) | 40 | 0.5 | Meningitis, bacteremia, UTI, pneumonia, osteomyelitis, and septic arthritis |
| Lejarzegi et al. (58) | Spain | Single | Prospective | 2,850 | 0–3 | 534 (22.2) | 1,207 (42.3) | 60 | 20 | Meningitis, bacteremia, and UTI |
| Lee et al. (59) | South Korea | Single | Retrospective | 145 | 0–3 | 88 (60.7) | 54 (37.2) | 5, 10, 15, 20 | N.A. | Meningitis only |
| Guo et al. (60) | China | Single | Retrospective | 498 | 0–3 | 32 (6.4) | 171 (34.3) | 53.8 | N.A. | Bacteremia only |
| Chang et al. () | Singapore | Single | Prospective | 187 | 0–3 | 54 (28.9) | 78 (41.7) | 7.2 | N.A. | Meningitis and bacteremia |
| Mahajan et al. (61) | USA | Multi | Prospective | 226 | 0–36 | 30 (13.3) | 131 (58.0) | N.A. | 0.5 | Meningitis, bacteremia, UTI, and pneumonia |
| Yin et al. (62) | China | Multi | Retrospective | 1,053 | 0–0.93 | 166 (15.8) | 432 (41.0) | N.A. | 2.4, 18 | Meningitis and bacteremia |
| Thongsamer et al. (63) | Thailand | Single | Retrospective | 134 | 3–36 | 29 (21.6) | 61 (45.5) | N.A. | 0.5 | Meningitis, bacteremia, UTI, and pneumonia |
| Kusma et al. (64) | USA | Single | Retrospective | 663 | 0.27–2 | 121 (18.3) | 287 (43.3) | N.A. | 0.5 | Meningitis, bacteremia, and UTI |
| Romain et al. (65) | France | Single | Retrospective | 357 | 0.14–1 | 45 (12.6) | 160 (44.8) | N.A. | 0.6, 2 | Meningitis, bacteremia, UTI, pneumonia, osteomyelitis, and septic arthritis |
| Kuppermann et al. (66) | USA | Multi | Prospective | 1,821 | 0–2 | 88 (9.6) | 765 (42.0) | N.A. | 0.5, 1.7 | Meningitis, bacteremia, and UTI |
Summary of studies included in the meta-analysis.
Diagnostic performance of C-reactive protein in detecting SBIs
The unadjusted and adjusted odds ratios for the detection of SBIs were 9.37 (95% CI: 5.12–17.1, I2 = 95.9%) and 3.91 (95% CI: 2.13–7.16, I2 = 96.7%), respectively, for CRP (Tables 2, 3, respectively). The unadjusted AUC for CRP in detecting SBIs was 0.81 (95% CI: 0.78–0.85, I2 = 76.4%) (Table 4). For the detection of SBIs, CRP with a cut-off value between 10 and 20 mg/L had the highest pooled sensitivity of 0.75 (95% CI: 0.54–0.89) and DOR of 13.84 (95% CI: 8.84–21.70) (Figure 2A, Table 5). CRP with a cut-off value of more than 40 mg/L had the highest pooled specificity 0.92 (95% CI: 0.87–0.95) and positive and negative likelihood ratios of 6.31 (95% CI: 4.15–9.59) and 0.54 (95% CI: 0.43–0.69), respectively (Figure 2A, Table 5).
Table 2
| Study | logOR | SE | Weight (common) (%) | Weight (random) (%) | Odds ratio IV, fixed + random, 95% CI | Odds ratio IV, fixed + random, 95% CI |
|---|---|---|---|---|---|---|
| Khoo, 2021 | 2.23 | 0.29 | 7.9 | 11.6 | 9.27 [5.27; 16.30] |  |
| Milcent, 2016 | 2.30 | 0.29 | 8.0 | 11.6 | 10.00 [5.70; 17.54] | |
| Leroy, 2018 | 4.12 | 0.12 | 44.2 | 12.6 | 61.60 [48.50; 78.24] | |
| Chiu, 2018 | 1.84 | 0.32 | 6.3 | 11.3 | 6.30 [3.35; 11.85] | |
| Galetto−Lacour, 2022 | 1.71 | 0.50 | 2.6 | 9.6 | 5.51 [2.06; 14.74] | |
| Yin, 2022 | 1.67 | 0.18 | 19.6 | 12.3 | 5.30 [3.70; 7.59] | |
| Rodriguez, 2022 | 0.97 | 0.44 | 3.4 | 10.2 | 2.63 [1.11; 6.23] | |
| Han, 2019 | 2.52 | 0.36 | 5.1 | 11.0 | 12.42 [6.16; 25.04] | |
| Li, 2022 | 2.44 | 0.48 | 2.9 | 9.8 | 11.53 [4.51; 29.48] | |
| Total (common effect, 95% CI) | 100.0 | 18.19 [15.52; 21.32] | ||||
| Total (random effect, 95% CI) | 100.0 | 9.37 [5.12; 17.14] | ||||
| Heterogeneity: Tau2 = 0.7370; Chi2 = 192.85, df = 8 (P < 0.0001); I2 = 95.9% Test for overall effect (random effects): Z = 7.26 (P < 0.0001) | ||||||
Forest plot of unadjusted odds ratio (OR) for C-reactive protein in detecting SBIs.
Table 3
| Study | logOR | SE | Weight (common) (%) | Weight (random) (%) | Odds ratio IV, fixed + random, 95% CI | Odds ratio IV, fixed + random, 95% CI |
|---|---|---|---|---|---|---|
| Khoo, 2021 | 2.10 | 0.34 | 0.0 | 11.1 | 8.16 [4.15; 16.04] |  |
| Milcent, 2016 | 1.44 | 0.35 | 0.0 | 11.1 | 4.20 [2.10; 8.40] | |
| Leroy, 2018 | 2.77 | 0.31 | 0.0 | 11.5 | 15.90 [8.70; 29.06] | |
| Chiu, 2018 | 0.02 | 0.00 | 99.5 | 12.9 | 1.02 [1.01; 1.03] | |
| Galetto−Lacour, 2022 | 1.72 | 0.71 | 0.0 | 7.7 | 5.59 [1.39; 22.48] | |
| Yin, 2022 | 1.61 | 0.18 | 0.1 | 12.4 | 5.00 [3.50; 7.14] | |
| Rodriguez, 2022 | 1.03 | 0.46 | 0.0 | 10.1 | 2.79 [1.14; 6.83] | |
| Han, 2019 | 1.67 | 0.42 | 0.0 | 10.5 | 5.33 [2.36; 12.04] | |
| Park, 2021 | 0.32 | 0.09 | 0.3 | 12.8 | 1.38 [1.16; 1.64] | |
| Total (common effect, 95% CI) | 100.0 | 1.02 [1.01; 1.03] | ||||
| Total (random effect, 95% CI) | 100.0 | 3.91 [2.13; 7.16] | ||||
| Heterogeneity: Tau2 = 0.7366; Chi2 = 245.70, df = 8 (P < 0.0001); I2 = 96.7% Test for overall effect (random effects): Z = 4.42 (P < 0.0001) | ||||||
Forest plot of adjusted odds ratio (OR) for C-reactive protein in detecting SBIs.
Table 4
| Study | MD | SE | Weight (common) (%) | Weight (random) (%) | Mean difference IV, fixed + random, 95% CI | Mean difference IV, fixed + random, 95% CI |
|---|---|---|---|---|---|---|
| Sutiman, 2022 | 0.74 | 0.04 | 5.9 | 7.3 | 0.74 [0.67; 0.81] |  |
| Hiremath, 2021 | 0.79 | 0.08 | 1.1 | 3.5 | 0.79 [0.63; 0.95] | |
| Nosrati, 2014 | 0.82 | 0.04 | 3.7 | 6.4 | 0.82 [0.73; 0.91] | |
| Manzano, 2011 | 0.88 | 0.02 | 17.7 | 8.8 | 0.88 [0.84; 0.92] | |
| Milcent, 2016 | 0.80 | 0.03 | 11.3 | 8.3 | 0.80 [0.75; 0.85] | |
| Diaz, 2016 | 0.68 | 0.04 | 4.4 | 6.8 | 0.68 [0.60; 0.76] | |
| Galetto−Lacour, 2022 | 0.75 | 0.06 | 1.9 | 4.8 | 0.75 [0.63; 0.88] | |
| Vujevic, 2017 | 0.83 | 0.03 | 7.9 | 7.8 | 0.83 [0.77; 0.89] | |
| Stein, 2014 | 0.60 | 0.11 | 0.6 | 2.3 | 0.60 [0.38; 0.82] | |
| Han, 2019 | 0.83 | 0.03 | 8.7 | 7.9 | 0.83 [0.77; 0.89] | |
| Markic, 2015 | 0.94 | 0.02 | 11.8 | 8.3 | 0.94 [0.90; 0.99] | |
| Parada, 2016 | 0.75 | 0.06 | 2.2 | 5.2 | 0.75 [0.64; 0.87] | |
| Hernandez−Bou, 2017 | 0.87 | 0.03 | 11.3 | 8.3 | 0.87 [0.82; 0.92] | |
| Kim, 2019 | 0.86 | 0.04 | 4.4 | 6.8 | 0.86 [0.78; 0.94] | |
| Lacroix, 2023 | 0.81 | 0.03 | 7.1 | 7.6 | 0.81 [0.74; 0.87] | |
| Total (common effect, 95% CI) | 100.0 | 0.84 [0.82; 0.85] | ||||
| Total (random effect, 95% CI) | 100.0 | 0.81 [0.78; 0.85] | ||||
| Heterogeneity: Tau2 = 0.0038; Chi2 = 59.21, df = 14 (P < 0.0001); I2 = 76.4% Test for overall effect (random effects): Z = 42.34 (P = 0) | ||||||
Forest plot of unadjusted area under the curve (AUC) for C-reactive protein in detecting SBIs.
Figure 2
Table 5
| Biomarker cut-offs | Sensitivity (95% CI) | Specificity (95% CI) | DOR (95% CI) | LR+ (95% CI) | LR− (95% CI) | AUC (95% CI) |
|---|---|---|---|---|---|---|
| C-reactive protein (mg/L) | ||||||
| <10 | 0.71 (0.60–0.80) | 0.64 (0.50–0.75) | 4.34 (3.55–5.31) | 1.96 (1.58–2.44) | 0.45 (0.37–0.55) | 0.79 (0.76–0.81) |
| 10–20 | 0.75 (0.54–0.89) | 0.82 (0.65–0.92) | 13.8 (8.84–21.7) | 4.20 (2.41–7.30) | 0.30 (0.17–0.54) | 0.84 (0.79–0.90) |
| 20–40 | 0.54 (0.42–0.65) | 0.88 (0.84–0.91) | 8.66 (6.13–12.2) | 4.54 (3.65–5.64) | 0.52 (0.42–0.65) | 0.78 (0.74–0.83) |
| >40 | 0.50 (0.37–0.63) | 0.92 (0.87–0.95) | 11.6 (7.089–19.1) | 6.31 (4.15–9.59) | 0.54 (0.43–0.69) | 0.77 (0.72–0.84) |
| Procalcitonin (ng/mL) | ||||||
| <0.5 | 0.78 (0.59–0.90) | 0.69 (0.54–0.81) | 8.02 (3.82–16.8) | 2.54 (1.73–3.72) | 0.32 (0.17–0.58) | 0.82 (0.73–0.93) |
| 0.5 | 0.64 (0.50–0.76) | 0.87 (0.81–0.91) | 12.1 (7.23–20.1) | 4.95 (3.59–6.83) | 0.41 (0.29–0.58) | 0.81 (0.76–0.86) |
| >0.5–2.0 | 0.58 (0.29–0.82) | 0.90 (0.79–0.96) | 12.7 (6.75–24.0) | 5.95 (3.71–9.56) | 0.47 (0.25–0.88) | 0.81 (0.74–0.90) |
Pooled diagnostic performance markers of C-reactive protein and procalcitonin for the detection of SBIs at multiple cut-offs.
A CRP cut-off between 10 and 20 mg/L showed the best diagnostic performance with a pooled AUC of 0.84 (95% CI: 0.79–0.90). This was followed by an AUC of 0.79 for CRP <10 mg/L (95% CI: 0.76–0.81), 0.78 (95% CI: 0.74–0.83) for CRP between 20 and 40 mg/L, and 0.77 (95% CI: 0.72–0.84) for CRP more than 40 mg/L (Figure 3A, Table 5). Using a data-driven approach, the continuous threshold for CRP that showed the best diagnostic performance was 28.7 mg/L (Table 6). However, the sensitivity for a CRP cut-off of 28.7 mg/L was 0.55 (95% CI: 0.41–0.67) compared with the categorical threshold of 10–20 mg/L, which had a sensitivity of 0.75 (95% CI: 0.54–0.89).
Figure 3
Table 6
| Biomarkers | Best cut-off | Sensitivity (95% CI) | Specificity (95% CI) |
|---|---|---|---|
| C-reactive protein (mg/L) | 28.707 | 0.55 (0.41, 0.67) | 0.86 (0.81, 0.90) |
| Procalcitonin (ng/mL) | 0.039 | 0.66 (0.51, 0.78) | 0.78 (0.70, 0.84) |
Best cut-off for C-reactive protein and procalcitonin as derived by Youden's index in detecting SBIs.
Overall, a CRP threshold of 10–20 mg/L demonstrated the highest pooled sensitivity, whereas a CRP level above 40 mg/L offered the highest pooled specificity.
Further sub-analyses were conducted to explore the robustness of biomarker performance across clinically relevant strata, including age group (infants aged 0–90 days) and SBI prevalence (low- vs. high-prevalence cohorts using 10% and 20% cut-offs). In the 0–90-day subgroup, pooled estimates for CRP and PCT thresholds showed trends consistent with the main analysis; however, no statistically significant differences in sensitivity or specificity were observed. Similarly, stratification by SBI prevalence also did not reveal meaningful differences in diagnostic performance between low- and high-prevalence studies, and heterogeneity remained substantial across strata. Given the absence of significant effect modification and the limited number of studies contributing to some subgroups, these analyses were considered exploratory. Full results of these sub-analyses are reported as Supplementary Figure.
Diagnostic performance of procalcitonin in detecting SBIs
For procalcitonin, the unadjusted and adjusted odds ratios were 4.62 (95% CI: 2.42–8.83, I2 = 97.3%) and 3.01 (95% CI: 1.60–5.65, I2 = 95.5%) (Tables 7 and 8, respectively), while the unadjusted AUC for procalcitonin was estimated to be 0.82 (95% CI: 0.73–0.93, I2 = 90.8%) (Table 9). Procalcitonin with a cut-off value of less than 0.5 ng/mL had the highest pooled sensitivity of 0.7812 (95% CI: 0.59–0.90) but the lowest pooled specificity of 0.69 (95% CI: 0.54–0.81) for the detection of SBIs (Figure 2B, Table 5). Procalcitonin between 0.5–2 ng/mL had the highest specificity at 0.90 (95% CI: 0.79–0.96) but the lowest sensitivity at 0.58 (95% CI: 0.29–0.82). At PCT between 0.5–2.0 ng/mL, the pooled DOR and positive and negative likelihood ratios were highest at 12.7 (95% CI: 6.75–24.0), 5.95 (95% CI: 3.71–9.56), and 0.47 (95% CI: 0.25–0.88), respectively (Figure 2B, Table 5).
Table 7
| Study | logOR | SE | Weight (common) (%) | Weight (random) (%) | Odds ratio IV, fixed + random, 95% CI | Odds ratio IV, fixed + random, 95% CI |
|---|---|---|---|---|---|---|
| Khoo, 2021 | 1.04 | 0.49 | 0.8 | 11.1 | 2.84 [1.08; 7.47] |  |
| Milcent (1), 2016 | 2.30 | 0.29 | 2.4 | 13.3 | 10.00 [5.70; 17.54] | |
| Leroy, 2018 | 1.74 | 0.12 | 13.5 | 14.5 | 5.70 [4.50; 7.22] | |
| Galetto−Lacour, 2022 | 1.30 | 0.50 | 0.8 | 11.1 | 3.67 [1.38; 9.76] | |
| Yin, 2022 | 1.69 | 0.18 | 6.1 | 14.2 | 5.40 [3.80; 7.67] | |
| Han (3), 2019 | 3.75 | 0.75 | 0.3 | 8.4 | 42.48 [9.68; 186.42] | |
| Kusma (CVI−), 2023 | 0.14 | 0.05 | 74.5 | 14.8 | 1.15 [1.04; 1.27] | |
| Kusma (CVI+), 2023 | 1.09 | 0.35 | 1.6 | 12.7 | 2.98 [1.50; 5.92] | |
| Total (common effect, 95% CI) | 100.0 | 1.73 [1.58; 1.88] | ||||
| Total (random effect, 95% CI) | 100.0 | 4.62 [2.42; 8.83] | ||||
| Heterogeneity: Tau2 = 0.7368; Chi2 = 262.51, df = 7 (P < 0.0001); I2 = 97.3% Test for overall effect (random effects): Z = 4.63 (P < 0.0001) | ||||||
Forest plot of unadjusted odds ratio (OR) for Procalcitonin in detecting SBIs.
Table 8
| Study | logOR | SE | Weight (common) (%) | Weight (random) (%) | Odds ratio IV, fixed + random, 95% CI | Odds ratio IV, fixed + random, 95% CI |
|---|---|---|---|---|---|---|
| Khoo, 2021 | 0.21 | 0.60 | 0.0 | 10.1 | 1.23 [0.38; 3.98] |  |
| Milcent (1), 2016 | 1.50 | 0.34 | 0.0 | 13.3 | 4.50 [2.30; 8.80] | |
| Leroy, 2018 | 1.50 | 0.26 | 0.0 | 14.2 | 4.50 [2.70; 7.50] | |
| Galetto−Lacour, 2022 | 1.78 | 0.73 | 0.0 | 8.6 | 5.94 [1.41; 25.02] | |
| Yin, 2022 | 1.67 | 0.18 | 0.1 | 14.9 | 5.30 [3.70; 7.59] | |
| Han (3), 2019 | 2.85 | 0.81 | 0.0 | 7.9 | 17.33 [3.54; 84.84] | |
| Park, 2021 | 0.25 | 0.10 | 0.2 | 15.4 | 1.29 [1.05; 1.58] | |
| Lejarzegi, 2023 | 0.01 | 0.01 | 99.6 | 15.6 | 1.01 [1.00; 1.02] | |
| Total (common effect, 95% CI) total (random effect, 95% CI) | 100.0 | 100.0 | 1.01 [1.00; 1.02] | |||
| Total (random effect, 95% CI) | 100.0 | 3.01 [1.60; 5.65] | ||||
| Heterogeneity: Tau2 = 0.6622; Chi2 = 156.94, df = 7 (P < 0.0001); I2 = 95.5% Test for overall effect (random effects): Z = 3.42 (P = 0.0006) | ||||||
Forest plot of adjusted odds ratio (OR) for Procalcitonin in detecting SBIs.
Table 9
| Study | MD | SE | Weight (common) (%) | Weight (random) (%) | Mean difference IV, fixed + random, 95% CI | Mean difference IV, fixed + random, 95% CI |
|---|---|---|---|---|---|---|
| Sutiman (1), 2022 | 0.63 | 0.04 | 3.4 | 5.8 | 0.63 [0.56; 0.71] |  |
| Hiremath, 2021 | 0.97 | 0.03 | 6.8 | 6.3 | 0.97 [0.92; 1.03] | |
| Mahajan (2), 2014 | 0.80 | 0.05 | 2.3 | 5.5 | 0.80 [0.71; 0.89] | |
| Manzano, 2011 | 0.82 | 0.03 | 7.6 | 6.3 | 0.82 [0.77; 0.87] | |
| Milcent (1), 2016 | 0.81 | 0.03 | 5.3 | 6.1 | 0.81 [0.75; 0.87] | |
| Diaz (IBI), 2016 | 0.77 | 0.10 | 0.5 | 3.2 | 0.77 [0.57; 0.97] | |
| Galetto−Lacour, 2022 | 0.78 | 0.06 | 1.6 | 5.1 | 0.78 [0.67; 0.89] | |
| Han (3), 2019 | 0.69 | 0.04 | 2.7 | 5.6 | 0.69 [0.61; 0.78] | |
| Stein, 2014 | 0.63 | 0.11 | 0.4 | 2.8 | 0.63 [0.41; 0.85] | |
| Markic (1), 2015 | 0.94 | 0.02 | 13.1 | 6.5 | 0.94 [0.90; 0.98] | |
| Thongsamer, 2022 | 0.80 | 0.05 | 2.3 | 5.5 | 0.80 [0.71; 0.89] | |
| Parada (1), 2016 | 0.65 | 0.06 | 1.2 | 4.7 | 0.65 [0.52; 0.77] | |
| Kusma (CVI−), 2023 | 0.88 | 0.03 | 7.6 | 6.3 | 0.88 [0.83; 0.93] | |
| Kusma (CVI+), 2023 | 0.79 | 0.05 | 2.3 | 5.5 | 0.79 [0.70; 0.88] | |
| Romain (SBI), 2022 | 0.87 | 0.03 | 4.8 | 6.1 | 0.87 [0.81; 0.93] | |
| Romain (IBI), 2022 | 0.97 | 0.01 | 30.1 | 6.7 | 0.97 [0.94; 1.00] | |
| Hernandez−Bou, 2017 | 0.88 | 0.04 | 3.9 | 5.9 | 0.88 [0.81; 0.95] | |
| Lacroix, 2023 | 0.77 | 0.03 | 4.1 | 6.0 | 0.77 [0.70; 0.84] | |
| Total (common effect, 95% CI) | 100.0 | 0.88 [0.87; 0.90] | ||||
| Total (random effect, 95% CI) | 100.0 | 0.82 [0.77; 0.86] | ||||
| Heterogeneity: Tau2 = 0.0090; Chi2 = 183.97, df = 17 (P < 0.0001); I2 = 90.8% Test for overall effect (random effects): Z = 32.97 (P < 0.0001) | ||||||
Forest plot of unadjusted area under the curve (AUC).
A PCT cut-off ≤0.5 ng/mL showed better performance based on SROC analysis (Figure 3B), although the pooled AUCs across the three cut-offs were similar (Table 5). Approaching procalcitonin as continuous data, the cut-off for procalcitonin that showed the best diagnostic performance was 0.039 ng/mL, which yielded a sensitivity of 0.66 (95% CI: 0.51–0.78) and a specificity of 0.78 (0.70–0.84).
Overall, PCT cut-off ≤0.5 ng/mL yielded the greatest pooled sensitivity but the lowest pooled specificity at 0.69 (95% CI: 0.54–0.81). Further sub-analysis could not be conducted for evaluation of PCT performance among different age groups and SBI prevalence in view of the small number of studies in each group.
Diagnostic performance of IL-6 in detecting SBIs
Only one study ultimately met the inclusion and exclusion criteria for IL-6 and no meta-analysis was performed. Notably, Zarkesh et al. () reported that IL-6 at a threshold of 20 pg/dL had a sensitivity and specificity of 79.1% and 91.6%, respectively. In the same study, CRP at a threshold of 10 mg/L had a lower specificity of 78.2% but higher sensitivity at 81.6%. Both CRP and IL-6 were reported to be of higher predictive accuracy than WBC count and ANC in predicting SBIs.
Risk-of-bias analysis
In this meta-analysis, Deeks' funnel plot asymmetry test was conducted to evaluate potential publication bias. No significant publication bias existed among the studies (P-value >0.2).
Discussion
The inclusion of biomarkers in clinical decision support tools in the evaluation and management of febrile children presenting with fever without source has in large part been supported by the increasing evidence of their diagnostic performances in distinguishing SBIs from self-limiting viral illnesses (, 67, 68). However, their optimal thresholds have remained inconclusive in view of high variability within and between studies. A recent systematic review and meta-analysis performed by Norman-Bruce et al. (69) evaluated the cut-offs of CRP and PCT for identification of invasive bacterial infections, but specifically only in febrile infants up to 90 days of life. With combined data from 37 unique data sets, this systematic review and meta-analysis is the latest and most comprehensive study to date to evaluate the thresholds at which the biomarkers CRP, PCT, and IL-6 can predict SBIs in children up to 36 months presenting with fever without source.
The results of this study suggest that CRP levels between 10 and 20 mg/L and PCT levels greater than 0.5 ng/mL have the best overall diagnostic accuracy in predicting SBIs in febrile children less than 36 months presenting with fever without source as reflected by their higher AUC values. These findings are in line with biomarker thresholds that have been validated in multiple clinical decision support tools such as Step-by-Step and AAP, both of which recommend cut-off values of 20 mg/L for CRP and 0.5 ng/mL for PCT (70, 71). We recognize that the high sensitivity of these thresholds is useful for ruling out SBIs, while moderate specificity may result in over-diagnosis and over-testing. Hence, these biomarkers have been incorporated in the algorithm in the aforementioned clinical decision support tools to limit unnecessary hospitalizations and invasive testing. Norman-Bruce et al. recently reported an optimum CRP level of 13.12 mg/L for the detection of invasive bacterial infections in febrile infants younger than 90 days and suggested the use of a lower cut-off of 15 mg/L.
The findings of this study also confirm that increasing the threshold value may increase specificity, enabling us to rule in the presence of SBIs with a greater degree of accuracy, with a CRP cut-off of 40 mg/L having the highest specificity of 0.92 (95% CI 0.87–0.95) and a PCT cut-off of above 0.5 ng/mL having the highest specificity of 0.90 (95% CI 0.79–0.96). Approaching CRP as a continuous variable, we determined the optimal threshold as 28.7 mg/L. However, this performed with a lower sensitivity compared with the categorical threshold of 10–20 mg/L. A higher threshold must be weighed against the risks of a higher false-negative rate, which may result in missed SBIs. Ultimately, the acceptable level of sensitivity and specificity and the decision on which threshold to use in clinical practice should also take into account heterogeneity of data and local prevalence of SBIs.
Indeed, there was high heterogeneity across studies, which may be due in part to differences in the definition of SBIs. The lack of a standardized definition of SBIs resulted in different infections with varying severity being identified across studies, which, in turn, led to variability in the outcomes assessed. This lack of standardization may result in either overestimation or underestimation of the association between biomarkers and SBI risk.
In addition, the majority of studies did not report the time at which the measurement of these biomarkers was performed. Given the pharmacokinetics of these biomarkers whereby CRP level reaches its maximum within 48 h, whereas PCT reaches its peak within 12 h (, , , ), this variability could introduce bias, particularly if some studies collect samples in the early stage of infection, while others do so in the later stage. Another factor contributing to heterogeneity is the age range differences between studies, particularly the distinction between neonates (0–3 months) and older infants (3–36 months). Younger children may present differently in terms of biomarker profiles and susceptibility to infections, influencing the biomarker thresholds used to identify SBIs. Furthermore, variability in laboratory methods, such as differences in assay techniques, sensitivity, and specificity, may lead to inconsistent results across studies. Finally, setting differences (e.g., emergency department vs. inpatient settings) may impact patient populations, with children in inpatient settings potentially reflecting more severe infections, thus affecting the generalizability of biomarker thresholds. Together, these factors contribute to heterogeneity by influencing both the detection of SBIs and the accuracy of biomarker thresholds, which may affect the overall precision of pooled estimates.
A majority of IL-6 studies were excluded primarily because they did not meet key eligibility criteria for this review, most commonly because of inappropriate age ranges, absence of confirmed SBI outcomes, or insufficient data to extract diagnostic accuracy measures or thresholds. As a result, only one study met the strict inclusion criteria, limiting the robustness of pooled estimates and preventing meaningful meta-analysis of IL-6 performance in this study. Although available data suggest potential diagnostic value, the lack of comparable studies precludes meaningful conclusions or recommendations for routine clinical use. Future research should prioritize standardization of IL-6 assays, including analytical methods, reference ranges, and reporting of thresholds. In addition, well-designed prospective studies should take into account the need to enrol a clearly defined population, apply consistent SBI definitions, report timing of biomarker sampling, and evaluate IL-6 as an independent marker alongside CRP and PCT.
This study has several strengths. First, we employed a comprehensive and rigorous search strategy, systematically screening four major databases with clearly defined inclusion and exclusion criteria. Our focus on young febrile children without a clear source of infection targets one of the most challenging clinical conundrums in evaluating febrile children up to 36 months of age. Second, we included only the most recent studies performed in the last 10 years to account for changing SBI epidemiology (72, 73). The inclusion of 37 studies in this study also allowed the aggregation of a large evidence base, hence strengthening the precision of pooled estimates. Methodologically, the use of hierarchical models, SROC curves, and diagnostic odds ratios ensured a robust synthesis of diagnostic performance of these biomarkers. In addition, the thresholds identified also showed clear alignment with existing clinical decision tools, including AAP and Step-by-Step approaches, supporting their applicability. We also provided transparent reporting of biomarker cut-offs and diagnostic accuracy metrics, with detailed tables and forest plots to facilitate interpretation by clinicians and researchers.
However, several limitations in this study must be acknowledged. Substantial heterogeneity across studies may weaken the reliability of pooled diagnostic estimates, due in part to the variability in SBI definitions and non-standardized timing of biomarker sampling, which represent a major potential source of bias. Evidence for IL-6 was extremely limited, with only one eligible study included. The dataset was also heavily weighted towards high-resource settings, which may limit generalizability to low- and middle-income contexts, and the exclusion of non-English publications may have introduced language bias. Finally, the predominance of retrospective study designs raises concerns about selection bias and incomplete ascertainment of SBI outcomes.
Conclusions
The findings of this review suggest that PCT at a threshold of 0.5 ng/mL and CRP at a threshold of 10–20 mg/L show the best diagnostic performance in identifying SBIs in febrile young children less than 36 months of age presenting with fever without source. These findings need to be validated in a large multi-centre study using specific timings of biomarker measurements and outcome definitions, as well as calibration in the local context before implementation.
Statements
Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material, and further inquiries can be directed to the corresponding author.
Author contributions
NS: Writing – original draft, Methodology, Data curation, Conceptualization, Writing – review & editing. JL: Writing – review & editing, Writing – original draft, Methodology, Data curation. RS: Software, Formal analysis, Writing – original draft, Methodology, Data curation, Investigation, Writing – review & editing. SY: Data curation, Methodology, Writing – review & editing, Writing – original draft. SG: Data curation, Writing – original draft, Methodology, Writing – review & editing. S-KR: Methodology, Writing – original draft, Data curation, Writing – review & editing. SG: Supervision, Writing – original draft, Data curation, Writing – review & editing, Methodology. S-LC: Conceptualization, Supervision, Investigation, Writing – review & editing, Methodology, Project administration, Funding acquisition, Data curation, Writing – original draft.
Funding
The author(s) declared that financial support was received for this work and/or its publication. The publication was funded by the National Medical Research Council of Singapore (MOH-001179-00). The funder had no role in the design or conduct of the study.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence, and reasonable efforts have been made to ensure accuracy, including review by the authors, wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fped.2026.1697210/full#supplementary-material
References
1.
SilvagniDCastagnoECardinaleFChiappaSChiarettiACorsiniIet alManagement of fever in infants and children in the emergency setting and during home discharge: recommendations from an Italian panel. Minerva Pediatr (Torino). (2022) 74:195–201. 10.23736/S2724-5276.22.06666-6
2.
KuzmanovićSRoncevićNStojadinovićA. Fever without a focus in children 0–36 months of age. Med Pregl. (2006) 59:187–91. 10.2298/MPNS0604187K
3.
Galetto-LacourAGervaixA. Identifying severe bacterial infection in children with fever without source. Expert Rev Anti Infect Ther. (2010) 8:1231–7. 10.1586/eri.10.118
4.
BassJWSteeleRWWittlerRRWeisseMEBellVHeisserAHet alAntimicrobial treatment of occult bacteremia: a multicenter cooperative study. Pediatr Infect Dis J. (1993) 12:466–73. 10.1097/00006454-199306000-00003
5.
BaraffLJBassJWFleisherGRKleinJOMcCrackenGHJrPowellKRet alPractice guideline for the management of infants and children 0–36 months of age with fever without source. Pediatrics. (1993) 92:1–12. 10.1542/peds.92.1.1
6.
ChangSSYLimAZOngGYKPiragasamRAllenJCNgKCet alPredictors of serious bacterial infections using serum biomarkers in an infant population aged 0–90 days: a prospective cohort study. BMJ Paediatr Open. (2021) 5:e000861. 10.1136/bmjpo-2020-000861
7.
TrippellaGGalliLDe MartinoMLisiCChiappiniE. Procalcitonin performance in detecting serious and invasive bacterial infections in children with fever without apparent source: a systematic review and meta-analysis. Expert Rev Anti-Infect Ther. (2017) 15:1041–57. 10.1080/14787210.2017.1400907
8.
ElmashadGMBahbahHMEl-NaidanySSESSobhSA. Assessment of serum interleukin.6 level in children with bacterial and viral meningitis. Menoufia Med J. (2021) 34:1358.
9.
ParanYYablecovitchDChoshenGZeitlinIRogowskiOBen-AmiRet alC-reactive protein velocity to distinguish febrile bacterial infections from non-bacterial febrile illnesses in the emergency department. Crit Care. (2009) 13:R50. 10.1186/cc7775
10.
Largman-ChalamishMWassermanASilbermanALevinsonTRitterOBerlinerSet alDifferentiating between bacterial and viral infections by estimated CRP velocity. PLoS One. (2022) 17:e0277401. 10.1371/journal.pone.0277401
11.
CosterDWassermanAFisherERogowskiOZeltserDShapiraIet alUsing the kinetics of C-reactive protein response to improve the differential diagnosis between acute bacterial and viral infections. Infection. (2020) 48:241–8. 10.1007/s15010-019-01383-6
12.
PaudelRDograPMontgomery-YatesAAYatacoAC. Procalcitonin: a promising tool or just another overhyped test?Int J Med Sci. (2020) 17:332. 10.7150/ijms.39367
13.
SamsudinIVasikaranSD. Clinical utility and measurement of procalcitonin. Clin Biochem Rev. (2017) 38:59.
14.
TanakaTNarazakiMKishimotoT. IL-6 in inflammation, immunity, and disease. Cold Spring Harb Perspect Biol. (2014) 6:a016295. 10.1101/cshperspect.a016295
15.
KlagTCantaraGSechtemUAthanasiadisA. Interleukin-6 kinetics can be useful for early treatment monitoring of severe bacterial sepsis and septic shock. Infect Dis Rep. (2016) 8:1–3. 10.4081/idr.2016.6213
16.
Buendia RodriguezJABedoyaMBenjumeaD. Usefulness of C-reactive protein and clinical characteristics in identifying severe bacterial infection in children with fever without Source. Turk Arch Pediatr. (2022) 57:616–20. 10.5152/TurkArchPediatr.2022.21348
17.
VijayanALVanimayaRavindranSSaikantRLakshmiSKartikRet alProcalcitonin: a promising diagnostic marker for sepsis and antibiotic therapy. J Intensive Care. (2017) 5:1–7. 10.1186/s40560-017-0246-8
18.
GomezBBressanSMintegiSDa DaltLBlazquezDOlacireguiIet alDiagnostic value of procalcitonin in well-appearing young febrile infants. Pediatrics. (2012) 130:815–22. 10.1542/peds.2011-3575
19.
MoherDShamseerLClarkeMGhersiDLiberatiAPetticrewMet alPreferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015 statement. Rev Esp Nutr Hum Diet. (2016) 20:148–60. 10.14306/renhyd.20.2.223
20.
SutimanNYaoSHWGohSSMSultanaRChongSL. Protocol for the diagnostic performance of C reactive protein, procalcitonin and interleukin-6 for serious bacterial infections among children ≤36 months old presenting with fever without source: a systematic review and meta-analysis. BMJ Paediatr Open. (2024) 8:e002237. 10.1136/bmjpo-2023-002237
21.
LinL. Bias caused by sampling error in meta-analysis with small sample sizes. PLoS One. (2018) 13:e0204056. 10.1371/journal.pone.0204056
22.
Veritas Health Innovation. Covidence—better systematic review management (2024). Available online at:https://www.covidence.org/(Accessed March 20, 2024).
23.
KoolMElshoutGMollHAKoesBWvan der WoudenJCBergerMYet alDuration of fever and course of symptoms in young febrile children presenting with uncomplicated illness. J Am Board Fam Med. (2013) 26:445–52. 10.3122/jabfm.2013.04.120265
24.
LeeCCChengJSChangYJChenYCHsinYCChiuCHet alSerious bacterial infections in young children with fever without source after discharge from emergency department: a national health insurance database cohort study. Pediatr Neonatol. (2022) 63:527–34. 10.1016/j.pedneo.2022.03.020
25.
BleekerSMoonsKDerksen-LubsenGGrobbeeDMollH. Predicting serious bacterial infection in young children with fever without apparent source. Acta Paediatr. (2001) 90:1226–31. 10.1111/j.1651-2227.2001.tb01566.x
26.
OuchenirLRenaudCKhanSBitnunABoisvertA-AMcDonaldJet alThe epidemiology, management, and outcomes of bacterial meningitis in infants. Pediatrics. (2017) 140:e20170476. 10.1542/peds.2017-0476
27.
RobertsKBDownsSMFinnellSMHellersteinSShortliffeLDWaldERet alUrinary tract infection: clinical practice guideline for the diagnosis and management of the initial UTI in febrile infants and children 2–24 months. Pediatrics. (2011) 128:595–610. 10.1542/peds.2011-1330
28.
FlorinTARamiloOHoyleJDJrAmbroggioLLorenzDShahSSet alRadiographic pneumonia in febrile infants 60 days and younger. Pediatr Emerg Care. (2021) 37:E221–6. 10.1097/PEC.0000000000002187
29.
ZainelAMitchellHSadaranganiM. Bacterial meningitis in children: neurological complications, associated risk factors, and prevention. Microorganisms. (2021) 9:1–12. 10.3390/microorganisms9030535
30.
YangBMallettSTakwoingiYDavenportCFHydeCJWhitingPFet alQuadas-C: a tool for assessing risk of bias in comparative diagnostic accuracy studies. Ann Intern Med. (2021) 174:1592–9. 10.7326/M21-2234
31.
RutterCMGatsonisCA. A hierarchical regression approach to meta-analysis of diagnostic test accuracy evaluations. Stat Med. (2001) 20:2865–84. 10.1002/sim.942
32.
TanBWongJJSultanaRKohJCJWJitMMokYHet alGlobal case-fatality rates in pediatric severe sepsis and septic shock: a systematic review and meta-analysis. JAMA Pediatr. (2019) 173(4):352–62. 10.1001/jamapediatrics.2018.4839
33.
SutimanNKhooZXOngGYKPiragasamRChongSL. Validation and comparison of the PECARN rule, step-by-step approach and lab-score for predicting serious and invasive bacterial infections in young febrile infants. Ann Acad Med Singap. (2022) 51:595–604. 10.47102/annals-acadmedsg.2022193
34.
HiremathPMNarayanappaDRajaniHS. C-Reactive protein with procalcitonin as a marker of occult bacterial infection in fever without a focus. J Nepal Paediatr Soc. (2021) 41(2):134–9. 10.3126/jnps.v41i2.31167
35.
NosratiABen TovAReifS. Diagnostic markers of serious bacterial infections in febrile infants younger than 90 days old. Pediatr Int. (2014) 56:47–52. 10.1111/ped.12191
36.
KhooZXChuaCYapZMKohJCChongSXWijedasaNHYet alA retrospective review of serious infections in febrile infants 0–90 days old. Clin Pediatr (Phila). (2021) 60(12–13):561–7. 10.1177/20101058211026003
37.
ManzanoSBaileyBGervaixACousineauJDelvinEGirodiasJBet alMarkers for bacterial infection in children with fever without source. Arch Dis Child. (2011) 96:440–6. 10.1136/adc.2010.203760
38.
MilcentKFaeschSGras-Le GuenCDubosFPoulalhonCBadierIet alUse of procalcitonin assays to predict serious bacterial infection in young febrile infants. JAMA Pediatr. (2016) 170:62–9. 10.1001/jamapediatrics.2015.3210
39.
DíazMGGarcíaRPGameroDBGonzález-ToméMIRomeroPCFerrerMMet alLack of accuracy of biomarkers and physical examination to detect bacterial infection in febrile infants. Pediatr Emerg Care. (2016) 32:664–8. 10.1097/PEC.0000000000000401
40.
LeeTGYuSTSoCH. Usefulness of the procalcitonin test in young febrile infants between 1 and 3 months of age. Korean J Pediatr. (2018) 61:285–90. 10.3345/kjp.2017.06170
41.
ChiuIMHuangLCChenILTangKSHuangYH. Diagnostic values of C-reactive protein and complete blood cell to identify invasive bacterial infection in young febrile infants. Pediatr Neonatol. (2019) 60:197–200. 10.1016/j.pedneo.2018.06.004
42.
D’SouzaAJShettyAKKD. Reactive thrombocytosis in febrile children with serious bacterial infection. J Evol Med Dent Sci. (2014) 3:12537–43. 10.14260/jemds/2014/3662
43.
Galetto-LacourACordeySPapisSMardeganCLuterbacherFCombescureCet alViremia as a predictor of absence of serious bacterial infection in children with fever without source. Eur J Pediatr. (2023) 182:941–7. 10.1007/s00431-022-04690-7
44.
HamielUBahatHKozerEHamielYZiv-BaranTGoldmanMet alDiagnostic markers of acute infections in infants aged 1 week to 3 months: a retrospective cohort study. BMJ Open. (2018) 8:e018092. 10.1136/bmjopen-2017-018092
45.
GomezBDiazHCarroABenitoJMintegiS. Performance of blood biomarkers to rule out invasive bacterial infection in febrile infants under 21 days old. Arch Dis Child. (2019) 104:547–51. 10.1136/archdischild-2018-315397
46.
VujevicMBenzonBMarkicJ. New prediction model for diagnosis of bacterial infection in febrile infants younger than 90 days. Turk J Pediatr. (2017) 59:261–8. 10.24953/turkjped.2017.03.005
47.
ZarkeshMSedaghatFHeidarzadehATabriziMBolooki-MoghadamKGhesmatiSet alDiagnostic value of IL-6, CRP, WBC, and absolute neutrophil count to predict serious bacterial infection in febrile infants. Acta Med Iran. (2015) 53:408–11.
48.
PolatEYücelH. Fever without source in children: six years experience of a children’s hospital. Gazi Med J. (2013) 24(4):148–52.
49.
HanSChoiSChoYS. Diagnostic markers of serious bacterial infections in infants aged 29–90 days. Signa Vitae. (2019) 15:22–6. 10.22514/SV151.042019.2
50.
MarkicJKovacevicTKrzeljVBosnjakNSapunarA. Lab-score is a valuable predictor of serious bacterial infection in infants admitted to hospital. Wien Klin Wochenschr. (2015) 127:942–7. 10.1007/s00508-015-0831-6
51.
MoldovanDABaghiuMDBalasATrutaST. The value of the “lab-score” method in identifying febrile infants at risk for serious bacterial infections. J Crit Care Med (Targu Mures). (2015) 1:11–7. 10.1515/jccm-2015-0003
52.
ParkJSByunYHLeeJYLeeJSRyuJMChoiSJet alClinical utility of procalcitonin in febrile infants younger than 3 months of age visiting a pediatric emergency room: a retrospective single-center study. BMC Pediatr. (2021) 21:109. 10.1186/s12887-021-02568-5
53.
ParadaECalaviaODurán-BallénMVasquezAAyatsRFerréNet alAssessing the use of procalcitonin in the hospitalised young febrile infant. An Pediatr (Barc). (2016) 84:278–85. 10.1016/j.anpedi.2015.08.011
54.
LiSYYinCHChenJSChenYSYangCCFangNW. A nomogram for predicting the development of serious bacterial infections in febrile term neonates: a single medical center experience in southern Taiwan. Pediatr Neonatol. (2022) 63:605–12. 10.1016/j.pedneo.2022.07.002
55.
Hernández-BouSTrenchsVVanegasMIVallsAFLuacesC. Evaluation of the bedside Quikread go® CRP test in the management of febrile infants at the emergency department. Eur J Clin Microbiol Infect Dis. (2017) 36:1205–11. 10.1007/s10096-017-2910-2
56.
KimJHLeeJYChoHRLeeJSRyuJMLeeJet alHigh concentration of C-reactive protein is associated with serious bacterial infection in previously healthy children aged 3–36 months with fever and extreme leukocytosis. Pediatr Emerg Care. (2019) 35:347–52. 10.1097/PEC.0000000000001080
57.
LacroixLPapisSMardeganCLuterbacherFL’HuillierASahyounC. Host biomarkers and combinatorial scores for the detection of serious and invasive bacterial infection in pediatric patients with fever without source. PLoS One. (2023) 18:e0294032. 10.1371/journal.pone.0294032
58.
LejarzegiABenitoJMintegiSGarcia-GarciaJJGomezBMartinez-VirumbralesLet alFebrile urinary tract infection in infants less than 3 months of age. Pediatr Infect Dis J. (2023) 42:E278–82. 10.1097/INF.0000000000003947
59.
LeeTGYuSTSoCH. Predictive value of C-reactive protein for the diagnosis of meningitis in febrile infants under 3 months of age in the emergency department. Yeungnam Univ J Med. (2020) 37:106–11. 10.12701/yujm.2019.00402
60.
GuoBCZhangXChenYWangYLiuGZhangHet alPredictors of bacteremia in febrile infants under 3 months old in the pediatric emergency department. BMC Pediatr. (2023) 23:444. 10.1186/s12887-023-04271-z
61.
MahajanPGrzybowskiMChenXKannikeswaranNStanleyRSingalB. Procalcitonin as a marker of serious bacterial infections in febrile children younger than 3 years old. Acad Emerg Med. (2014) 21:171–9. 10.1111/acem.12316
62.
YinZWuQZhangYChenJLiuCLiJet alA novel algorithm with paired predictive indexes to stratify the risk levels of neonates with invasive bacterial infections: a multicenter cohort study. Pediatr Infect Dis J. (2022) 41:E149–55. 10.1097/INF.0000000000003437
63.
ThongsamerPLaoprasopwattanaK. Using procalcitonin levels to predict infection and reduce unnecessary antibiotic usage in febrile children aged 3–36 months. J Infect Dev Ctries. (2022) 16:1445–9. 10.3855/jidc.14640
64.
KusmaJFuQKafleMLibrizziJMireaLRigginsH. Effect of viral illness on procalcitonin as a predictor of bacterial infection in febrile infants. Hosp Pediatr. (2023) 13:961–6. 10.1542/hpeds.2022-007070
65.
RomainASBasmaciRKwonTGendrelDRaymondJChalumeauMet alProcalcitonin at 12–36 h of fever for prediction of invasive bacterial infections in hospitalized febrile neonates. Front Pediatr. (2022) 10:968207. 10.3389/fped.2022.968207
66.
KuppermannNDayanPSLevineDAVitaleMTzimenatosLJaffeDMet alA clinical prediction rule to identify febrile infants 60 days and younger at low risk for serious bacterial infections. JAMA Pediatr. (2019) 173:342–51. 10.1001/jamapediatrics.2018.5501
67.
AronsonPLNeumanMI. Should we evaluate febrile young infants step-by-step in the emergency department?Pediatrics. (2016) 138(3):e20161579. 10.1542/peds.2016-1579
68.
VelascoRGomezBBenitoJMintegiS. Accuracy of PECARN rule for predicting serious bacterial infection in infants with fever without a source. Arch Dis Child. (2021) 106(2):143–8. 10.1136/archdischild-2020-318882
69.
Norman-BruceHUmanaEMillsCMitchellHMcFetridgeLMcClearyDet alDiagnostic test accuracy of procalcitonin and C-reactive protein for predicting invasive and serious bacterial infections in young febrile infants: a systematic review and meta-analysis. Lancet Child Adolesc Health. (2024) 8:358–68. 10.1016/S2352-4642(24)00021-X
70.
PantellRHRobertsKBAdamsWGDreyerBPKuppermannNO’LearySTet alEvaluation and management of well-appearing febrile infants 8–60 days old. Pediatrics. (2021) 148(2):e2021052228. 10.1542/peds.2021-052228
71.
GomezBMintegiSBressanSDa DaltLGervaixALacroixLet alValidation of the “step-by-step” approach in the management of young febrile infants. Pediatrics. (2016) 138(2):e20154381. 10.1542/peds.2015-4381
72.
BiondiEALeeBRalstonSLWinikorJMLynnJFDixonAet alPrevalence of bacteremia and bacterial meningitis in febrile neonates and infants in the second month of life: a systematic review and meta-analysis. JAMA Netw Open. (2019) 2:e190874. 10.1001/jamanetworkopen.2019.0874
73.
GreenhowTLHungYYHerzAMLosadaEPantellRH. The changing epidemiology of serious bacterial infections in young infants. Pediatr Infect Dis J. (2014) 33:595–9. 10.1097/INF.0000000000000225
Summary
Keywords
bacterial, biomarker, children, infection, threshold
Citation
Sutiman N, Lin J, Sultana R, Yao SHW, Goh SSM, Rocknathan S-K, Ganapathy S and Chong S-L (2026) Identifying the thresholds of C-reactive protein, procalcitonin, and interleukin-6 among children ≤36 months’ old with fever without source at risk of serious bacterial infections: a systematic review and meta-analysis. Front. Pediatr. 14:1697210. doi: 10.3389/fped.2026.1697210
Received
09 September 2025
Revised
23 December 2025
Accepted
02 January 2026
Published
26 February 2026
Volume
14 - 2026
Edited by
Nahed Ismail, University of Illinois Chicago, United States
Reviewed by
Yara Algoraini, King Fahad Medical City, Saudi Arabia
Priyam Pattnaik, University of Connecticut Health Center, United States
Updates
Copyright
© 2026 Sutiman, Lin, Sultana, Yao, Goh, Rocknathan, Ganapathy and Chong.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Shu-Ling Chong gmscshul@nus.edu.sg
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.